Make test suite runnable from clean checkout + cross-platform vram_detect

runnable-test-suite (parent) — complete. Three sub-tasks all complete:

- make-tests-runnable: add requirements.txt pinning pytest==7.4.4, sweep all
  docs/prompts from bare 'python' to 'python3' (stock macOS/Windows ships
  python3), add idempotent .venv install block to scripts/install.sh, and add
  'from __future__ import annotations' to 3 dashboard modules using PEP 604
  union syntax at definition time so they import on Python 3.9+. The
  PEP 604 bug was caught by the streak verifier itself during implementation.

- vram-detect-cross-platform: scripts/vram_detect.py now branches on
  platform.system() for Linux/Darwin/Windows. macOS path uses
  system_profiler SPDisplaysDataType (Apple Silicon unified memory via
  sysctl hw.memsize; Intel Macs via 'VRAM (Total):'). Windows uses wmic
  path win32_VideoController get AdapterRAM with PowerShell fallback. Linux
  /proc/meminfo and nvidia-smi/lspci paths unchanged (regression test locks
  them). Added 14 local-LLM context-window entries (llama-3.1, qwen2.5,
  mistral, deepseek-r1/v3, glm-4/4.5, gemma-2, phi-3/4) with source-cited
  model cards. Added _probe_ollama_model() that runs 'ollama list' as a
  last-resort fallback. run_command() now wraps PowerShell cmdlets on
  Windows (['powershell', '-NoProfile', '-NoLogo', '-Command', ...]).

- vram-detect-cross-platform-tests: 11 new monkeypatched tests in
  tests/test_vram_detect.py covering Linux/Darwin/Windows branches for
  detect_ram and detect_gpu_vram, prefix-match for unknown model names,
  ollama probe, Windows PowerShell wrapper, and a LOCKED regression test
  for _detect_ram_linux(). All external subprocess/sysctl/wmic calls are
  mocked; no live hardware probes. Suite total: 235 passed, 0 errors.

Verified on this box: gpu_vram_gb 0 -> 32 on Apple M5 (32GB unified memory),
target context correctly jumped 12k -> 42k.

Subtask-2 implementation was authored by local LLM (gemma-4-26B-A4B-it
via headroom proxy @ localhost:8787). The 10-consecutive-clean-pass streak
verifier ran as the independent checker model (article #2/#9/#13 in
'WTF Is a Loop? Part 2'). One anti-spin rail fired: local LLM produced
inline branches where subtask-3 tests expected private _detect_ram_linux()
helper; extracted helper to match the test contract without weakening tests.

Parent + all 3 subtasks complete. Prior opencode-subagent implementation
of subtask-2 preserved in git stash for reference.

tasks/runnable-test-suite/subtasks/vram-detect-cross-platform-tests/SPEC.md

@@ -0,0 +1,126 @@
+# SPEC — vram-detect-cross-platform-tests
+
+Parent: `runnable-test-suite` (see PARENT_SPEC.md).
+
+**Dependency:** This sub-task runs AFTER `vram-detect-cross-platform` is complete.
+Its function signatures and detection logic are the contract you test against.
+
+## Scope
+
+pytest tests proving cross-platform branches of `scripts/vram_detect.py` work
+WITHOUT hitting real hardware. Every `subprocess.run` / `Path.exists` / `sysctl`
+call is mocked.
+
+## Files this sub-task touches (and ONLY this)
+
+- `tests/test_vram_detect.py` — extend or rewrite. Preserve any existing
+  passing test (Linux regression tests especially).
+
+## MUST NOT touch
+
+- `scripts/vram_detect.py` (if it has a bug, escalate back to Orchestrator —
+  subtask-2 owns it)
+- Any documentation, prompt, install script, or other source file
+- Any other test file
+
+## Test patterns (parametrize + monkeypatch)
+
+### 1. `test_detect_ram_linux` (regression — must already be there, keep it)
+- Patches `Path("/proc/meminfo")` text with synthetic `MemTotal: 16384000 kB\nMemAvailable: 8192000 kB`.
+- Asserts `(16_384_000, 8_192_000)` returned.
+
+### 2. `test_detect_ram_macos`
+- `monkeypatch.setattr(platform, "system", lambda: "Darwin")`.
+- Patches `subprocess.run` so `sysctl -n hw.memsize` returns `"34359738368"` (32GB).
+- Asserts total_kb correct (33_554_432), available == total.
+
+### 3. `test_detect_ram_windows`
+- `monkeypatch.setattr(platform, "system", lambda: "Windows")`.
+- Patches subprocess to return wmic output `TotalPhysicalMemory=34359738368`.
+- Asserts total_kb == 33_554_432.
+
+### 4. `test_detect_gpu_vram_nvidia_linux`
+- `monkeypatch.setattr(platform, "system", lambda: "Linux")`.
+- Patches `nvidia-smi` output `"24576\n"`.
+- Asserts `(25_165_824, 25_165_824, 1)` (24GB × 1024 = KB).
+
+### 5. `test_detect_gpu_vram_apple_silicon`
+- `monkeypatch.setattr(platform, "system", lambda: "Darwin")`.
+- Patches `system_profiler SPDisplaysDataType` with a real Apple M-series snippet:
+  ```
+  Graphics/Displays:
+      Apple M2:
+        Chipset Model: Apple M2
+        Type: GPU
+        Bus: Built-In
+        Total Number of Cores: 10
+        Vendor: Apple (0x106b)
+        Metal: Supported, version 2
+  ```
+- Patches `sysctl -n hw.memsize` with `"17179869184"` (16GB).
+- Asserts gpu_vram_kb > 0, num_gpus >= 1.
+
+### 6. `test_detect_gpu_vram_windows_wmic`
+- `monkeypatch.setattr(platform, "system", lambda: "Windows")`.
+- Patches wmic output:
+  ```
+  AdapterRAM=8589934592
+  Name=NVIDIA GeForce RTX 3060
+  ```
+- Asserts total_vram_kb == 8_388_608 (8GB).
+
+### 7. `test_lookup_model_context_unknown_returns_zero`
+- `_lookup_model_context("completely-unknown-model")` returns `0`.
+
+### 8. `test_lookup_model_context_prefix_match`
+- `_lookup_model_context("deepseek-r1:7b")` matches `deepseek-r1` entry and returns its context.
+- `_lookup_model_context("llama-3.1-8b-instruct")` matches `llama-3.1-8b`.
+
+### 9. `test_detect_model_context_ollama_probe`
+- No config file specifies a model.
+- `ollama list` is on PATH (patch `shutil.which`).
+- Patches subprocess to return:
+  ```
+  NAME           ID     SIZE    MODIFIED
+  llama-3.1-8b   abc    4.7GB   2 days ago
+  ```
+- Asserts the returned context matches the `llama-3.1-8b` entry.
+
+### 10. `test_run_command_windows_powershell_wrapper`
+- Show that on Windows, a PowerShell cmdlet invocation goes through
+  `powershell -NoProfile -NoLogo -Command "..."` (assert argv[0] is powershell
+  when cmd is a PS string).
+
+### 11. `test_detect_ram_linux_regression` (LOCKED — do not modify)
+- If this test already exists, keep its byte-for-byte assertions. If not, add
+  an exact match on `(total_kb, available_kb)` for a specific `/proc/meminfo`
+  fixture to prevent subtask-2 from regressing Linux output.
+
+## Acceptance criteria
+
+1. `python3 -m pytest tests/test_vram_detect.py -v` exits 0 with all new tests passing.
+2. Total `python3 -m pytest tests/ -v` still exits 0.
+3. **Streak:** 10 consecutive clean `python3 -m pytest tests/ -v` runs.
+4. No live `subprocess` against real `system_profiler`/`nvidia-smi`/`wmic` — every
+   external call goes through `monkeypatch.setattr`.
+5. Every new test runs < 500ms (mock only, no I/O).
+
+## Anti-spin rails
+
+- If a subtask-2 function signature doesn't support what a test needs, STOP and
+  report. Do not edit `vram_detect.py` yourself and do not weaken the test to fit.
+  Escalate via Orchestrator.
+- No `pytest.mark.skip` unless the platform genuinely doesn't support the feature
+  (e.g. skip a Windows test on Linux only if it can't be mocked — but mocking is
+  the whole point, so this should never happen).
+- Do not add `pytest-cov` or any new deps.
+
+## Recommended approach
+
+1. Read `scripts/vram_detect.py` and confirm subtask-2's signatures.
+2. For each test, write the fixture data as a module-level constant (multiline string).
+3. Use `monkeypatch.setattr` for `platform.system`, `subprocess.run`, `shutil.which`,
+   `Path.exists`, `Path.read_text`. Never call the real thing.
+4. Use `capfd` for stdout assertions where the SPEC calls for print messages.
+5. Run `python3 -m pytest tests/test_vram_detect.py -v` until green.
+6. Run the full suite streak verifier.