Adopt research-repo-tools v0.1.9 performance tooling before v0.5.0 #276

Summary

Adopt the published research-repo-tools v0.1.9 capabilities from acgetchell/research-repo-tools#94–#98 before la-stack v0.5.0. Replace the corresponding generic Python and workflow logic with supported shared APIs and configuration under tooling/, while preserving scientific eligibility and the dimension plot.

Current State

PR #275 merged the exact published v0.1.8 dependency and completed #268 and #273. Common-harness measurement, complete immutable evidence and dependency-body preservation already use shared functionality. Those completed migrations remain closed.

Remaining generic code exists because v0.1.8 does not yet provide all the required plotting, provenance, complete-run document publication, Criterion verification and release-target handoff contracts. Full-report standardization against the existing shared format can proceed independently; new v0.1.9 APIs must wait for publication.

Proposed Changes

• Wait for Publish research-repo-tools v0.1.9 performance workflow extensions research-repo-tools#99 to verify the published v0.1.9 wheel and sdist. Pin the exact registry release with required extras and refresh locks through the supported update workflow (just update); do not substitute a sibling checkout or unreleased wheel.
• Inventory affected paths, tests and recipes for each capability, then migrate generic behavior and regressions upstream. Retain focused installed-package integrations and domain assertions.
• Standardize the full report and README publication on complete-run evidence, with declarations in tooling/ and thin canonical recipes. Keep the dimension plot.
• Preserve existing published report/artifact bytes and archives. Do not automatically prune historical reports. Retire a legacy regeneration path only after its supported replacement and retirement condition are documented; never fabricate complete-run evidence from summary CSVs.
• Record before/after production Python lines, test Python lines and collected test counts, with a disposition for each retained helper or test group.
• Run the repository's required focused checks and installed-package workflow tests. Verify required Linux, macOS and Windows consumer CI for the actual adoption commit before closing, including applicable native command discovery, executable invocation, path/encoding and subprocess failure behavior. Link commit/run evidence; a macOS-only pass or older PR head is insufficient.
• Merge the adoption and update ownership/benchmark/release guidance before v0.5.0. Generate any new v0.5.0 performance evidence only through the validated adopted workflow; do not claim new measurements from fixture tests.

Benefits

Minimizes local Python, centralizes reusable regressions and leaves la-stack responsible for scientific policy, configuration and release evidence.

Implementation Notes

Native prerequisite: acgetchell/research-repo-tools#99 (v0.1.9 publication). This issue blocks completion of acgetchell/research-repo-tools#100, the separate v0.1.9 adoption tracker, and belongs to milestone v0.5.0. It does not reopen #268/#273 or make the older acgetchell/research-repo-tools#69 rollout a prerequisite.

Do not weaken scientific gates, release identity checks, historical evidence provenance or supported Windows behavior to remove code. Record any capability deferred upstream as unfinished adoption with an explicit replacement issue and release decision.