feat(quality): add dynamic test quality runners

[vicotrbb]

Closes #10.

Scope

Semantic owner: Quality Engineering.

This implements the EG-DYNAMIC descriptor registry and its public cargo xtask quality / immutable evidence-report seam. It adds bounded target identities, retained corpus/seed/schedule/minimized-failure/output-protocol data, controlled target execution, and fail-closed cancellation, timeout, malformed-output, bounded-capture, missing-tool, stale-registry, and no-retry behavior. It also adds a focused EG-COVERAGE detector-availability contract audit without changing coverage policy or production coverage commands.

Qualification status

Implemented, not Qualified. This change makes no Qualification Cell, merge-eligibility, approval, or release-readiness claim.

Evidence

Focused checks passed:

• cargo test --locked --package xtask --test foundational_scope_activation -- m0_09_ — 15 tests
• cargo test --locked --package xtask dynamic_quality — registry unit coverage
• cargo fmt --check
• cargo clippy --locked --package xtask --all-targets --all-features -- -D warnings
• policy JSON validation and git diff --check
• clean isolated cargo xtask quality --profile pr for ea26c13; EG-DYNAMIC and EG-TEST passed.

EG-COVERAGE is intentionally not selected by the PR profile. Its contract is covered through existing public EXT/focused tests plus the new detector-availability fixture; no coverage threshold, selection, exception, or command changed.

Preserved diagnostic failures were not erased: an initial root-target PR attempt failed EG-EVIDENCE on historical retained evidence, and an isolated b05a142 attempt failed EG-TEST because its summary parser classified 3 slow as noncanonical skips. The final exact ea26c13 isolated PR-profile attempt passed.

[vicotrbb]

M0-09 final review remediation is pushed at exact head 4637e3bd9d8260bfebef6d4509fd5d1a93b4659e.

Implemented narrowly:

• frozen seven-kind dynamic capability catalog, separate from the two active M0 targets;
• independent retained-evidence verifier with its own grammar parsing and target/plan/tool/version/argv/environment/input/catalog/registry/result recomputation;
• fail-closed copied-source tamper coverage for plan, command, environment, result, and target order/cardinality;
• EG-DYNAMIC remains PR/EXT only and is absent from QUAL execution;
• canonical Nextest (N slow) passing summaries are accepted while malformed, skipped, and failure-bearing summaries remain rejected.

Focused evidence:

• M0-09 public-seam suite: 22/22 passed;
• independent verifier subset: 4/4 passed;
• dynamic unit tests: 2/2 passed;
• focused llvm-cov smoke passed;
• fmt, clippy, JSON, and diff checks passed.

Fresh unprepared exact-head push from a new detached worktree and new external Cargo target passed every authoritative PR gate: 501/501 Nextest tests passed (2 canonical slow annotations), 16 doctests passed, EG-DYNAMIC/EG-EVIDENCE/EG-TEST passed. Diagnostic evidence: /private/tmp/positron-m0-09-final-push-worktree.9kanE5/target/quality/evidence/1785400960192-4637e3bd9d82-71800.json (local diagnostic, not a trusted-CI merge claim).

Preserved failed attempts: the shared worktree first failed stale historical EG-EVIDENCE plus two unrelated M0-08 reap timing tests; the first isolated run passed 501/501 but exposed the pre-existing slow-summary parser regression. Neither failure was erased. GitHub CI run 30528207118 is now running for the exact pushed head.