[kvlonge]

Why

Remove avoidable Python dependencies from caching, compile reuse and runtime bookkeeping while preserving supported Python execution and APIs.

Changes

• Move version-5 SQL-test snapshots into the cache crate, preserving exact bytes and malformed-cache misses.
• Assemble identity payloads, CTE arrays and sorted pairs as Rust JSON.
• Keep compile-reuse checks, launch records, prepared SQL-test plans and managed paths native; preserve public Python rows and path semantics.
• Convert SQL-test, audit and scenario results once for summaries, failures and JSON, including history-projection metadata; preserve opaque SQL cells and callback artifacts.
• Use the existing bundled SQLite library for history connections and an interruptible native lock. Preserve connection/subclass hooks, recovery, literal filenames and Ctrl+C.
• Delete replaced helpers and lower boundary budgets without new exceptions or warehouse-state changes.

Verification

• 45 cache and 9 binding Rust tests; 53 cache/result, 65 reuse/launcher, 147 artifact-planning, 110 identity, 12 audit/report, 34 scenario/report and 162 history tests passed.
• make test: 12,185 passed after artifact-plan and identity migrations.
• Exact make check-ci, bounded independent reviews and private hygiene scans passed.
• Cold/warm launcher, exact cache/identity/artifact parity and unchanged warm mtimes verified.
• Scenario strings match baseline bytes; direct/local/replay reports match after timing/path normalization.
• SQLite works without Python connect. Three paired runs over 8,000 events/2,000 runs match exactly; peak RSS rises about 9 MiB. Literal-path and SIGINT regressions covered.
• No external real project was available; no speedup claim. Windows source-reviewed, not run locally; SIGINT subprocess probe is POSIX-only.
• No DuckShop addition: behavior-preserving boundary refactor, not a project-shape change.