Review and integrate v0.2.7 validation parallelism and Cargo build caching #338

Integrate the shared validation-speed improvements delivered by merged research-repo-tools PR #207, targeting v0.2.7: parallel independent validation stages and reusable Cargo compilation output with separate managed installation directories.

The maintainer authorizes exact development revision cd8594f658b5bf3ade063a685834565cc62e0cb3 for implementation in PR #339. This issue tracks delivery of the integration and closes when that PR merges. #333 owns verified supplier publication, normal published-package adoption, final native validation and matched cold/warm timings. This handoff supersedes older release-only acceptance here; a development pin is not publication evidence.

Repository integration

Review independent Python/repository checks versus Rust tests, documentation, exact-package checks and benchmark compilation. Keep measured benchmark acquisition isolated from concurrent validation. Coordinate #326 and #333 for final published-tooling adoption, and #330 for release-flow parity; this issue owns the validation-speed integration.

Acceptance

• Record the current command/dependency graph and representative local and native-CI stage timings. Map every existing validator to its final owner; distinguish queue time, execution time and total runner usage.
• Integrate the maintainer-authorized exact development revision consistently in the manifest, lock and isolated recipe/workflow invocations. Use the supplier's public APIs and supported cache boundary rather than copying a cache or orchestration engine. release: plan, publish and verify la-stack v0.5.3 #333 owns replacement with the verified published package and final released-API adoption checks.
• After fast policy checks, overlap only independent source, installed-package, setup or documentation stages that this repository actually has. Keep build/generation prerequisites ordered and serialize operations sharing mutable environments or output paths. Retain a complete local just ci and document its serial fallback.
• For managed Cargo tools, use the shared build-cache default or documented absolute RESEARCH_REPO_TOOLS_CARGO_BUILD_HOME override. Preserve separate exact-version installation roots, locked builds and executable verification. CI caches contain compilation output only, use OS/architecture and toolchain/tool declaration keys, and respect branch trust boundaries. Preserve a fresh-install smoke check where one already exists.
• Preserve required-check/provider names where practical. If job ownership changes, reconcile branch protection declarations, publishing/evidence policy and required job/step identities together. Missing, failed, cancelled or skipped required stages must not become success. Require all native platforms already supported.
• Run focused regressions and final local checks for the implementation, and record the observed timings, justified non-applicability and remaining bottlenecks. Required PR checks still gate merge. release: plan, publish and verify la-stack v0.5.3 #333 owns final release-revision native validation and matched cold-cache/warm-cache measurements; compare equivalent coverage and workload before claiming a speedup.
• Update contributor/validation guidance and link the implementation PR and measured results here. Request Astra review if the integration changes evidence ownership, signing/approval, retry/recovery or scientific-comparability contracts.

Implementation and measured local results are recorded in PR #339 and its integration report. Both serial and parallel full local CI passed with 438 Python and 1,143 Rust tests; their different compilation-cache states do not establish an isolated speedup. The PR carries Closes #338; publication and release validation remain pending under #333.

This issue tracks integration work. It does not authorize changing scientific workloads, weakening checks, starting benchmark campaigns, deploying, tagging or publishing releases. Existing release-order and milestone requirements remain in force.