Preserve failure diagnostics and classify unavailable capture prerequisites

[morluto]

Summary

This PR addresses #87–#92 as one failure-observability and environment-readiness change:

• preserves managed Perfetto staging phase, category, and bounded cause in receipts and durable operation status;
• reports denied perf sampling as unusable, including the observed perf_event_paranoid restriction and remediation;
• adds bounded active CUDA toolkit preflight for declared nvcc workloads and classifies missing development headers as environment_blocked;
• preserves Torch-profiler wrapper phase, diagnostics metadata, stdout/stderr, and collector failure limitations;
• classifies external compiler failures in pytest evidence without counting them as application failures;
• allows MCP startup and inspection to bind an explicit external workspace while retaining the project root for workloads and source provenance.

Root cause

The affected paths generally retained only a terminal error or binary-presence check. Managed staging replaced underlying network/filesystem/verification causes, operation progress never received the staging phase, capability probes treated perf --version as sufficient, and adapter/test evidence lacked a bounded way to distinguish wrapper or toolchain failures from workload defects. MCP initialization also did not pass its existing external workspace capability through the server boundary.

Validation

• uv run pytest -q — 299 passed, 112 deselected
• uv run ruff check src tests
• uv run ruff format --check src tests
• uv run mypy src tests
• git diff --check

The deselected tests are optional/provider-specific integration lanes; synthetic regression tests cover the new failure paths, while a real Torch/GPU environment was not available for those optional lanes.

Closes #87
Closes #88
Closes #89
Closes #90
Closes #91
Closes #92

[cursor]

Bugbot is not enabled for your account, so this pull request was not reviewed.

Enable Bugbot in the Cursor dashboard to get automatic reviews on future PRs.

[chatgpt-codex-connector]

💡 Codex Review

Here are some automated review suggestions for this pull request.

Reviewed commit: b145dee99d

ℹ️ About Codex in GitHub

Your team has set up Codex to review pull requests in this repo. Reviews are triggered when you

• Open a pull request for review
• Mark a draft as ready
• Comment "@codex review".

If Codex has suggestions, it will comment; otherwise it will react with 👍.

Codex can also answer questions or update the PR. Try commenting "@codex address that feedback".