Validate all accepted runtime improvements

[KulmanD]

What changed

• combine accepted Issues Fail safely when an event tap interrupts an active gesture #21, Abort stale gesture holds without timers #23, Reset scroll smoothing state between independent scroll sequences #24, and Preserve all CGEvent scroll delta representations #25 directly on the frozen baseline
• exclude rejected movement-delta Issue Use CGEvent movement deltas for gesture accumulation #22
• retain isolated, rejected, pairwise, and cumulative performance reports
• add cumulative tests for stale/interrupted gesture recovery handing scroll back to the transformed inactive path
• optimize one-time event-tap recovery diagnostics to stay below a Mach-O text-page boundary
• include the development-only repeated measurement runner

Why

The final runtime set needs a cumulative interaction gate. The direct composition crossed a 16 KiB __TEXT boundary despite adding little actual code; the optimized composition restores the smaller image without changing worker behavior.

Validation

• swift build
• swift build -c release
• GitHub Actions: 56 tests, 0 failures before the documentation-only evidence import
• helper contains no XCTest/Testing linkage
• app and lightweight source archives verify and exclude tests/performance probes
• Start/Status/Stop leaves no process
• user config hash unchanged
• no timer, polling, persistent thread, event history, network path, normal log, or reposted scroll event

Performance and remaining checks

Direct composition: 450616 bytes with a 229376-byte __TEXT segment.

Optimized composition: 433528 bytes with the baseline 212992-byte __TEXT segment, only 976 bytes (+0.226%) over the frozen helper.

Long repeated idle, combined active workloads, and physical MX Master/trackpad validation are deliberately delegated:

bash scripts/performance/measure-candidate.sh --label all-runtime-integration

This PR remains draft until those results are recorded. Related: #21, #23, #24, #25.

[KulmanD]

Superseded by draft PR #40, which contains this change, its performance evidence, the complete accepted integration, and passing final CI. Closing this draft without merging avoids duplicate commits. The branch is intentionally retained for isolated comparison and deferred measurements.