fix(dev-proxy): harden backend restarts and cleanup

[debugmcpdev]

Summary

• let both restart tools replace, preserve, or clear persistent backend environment overrides without passing them to builds
• report display-safe overrides in status using the shared sensitive-name and credential-shape redactors, with a fail-closed bootstrap mode and proxy-only opt-out
• label custom docker run backends and remove owned containers for stdio, HTTP, and SSE lifecycle paths
• serialize automatic startup, restarts, and shutdown so an immediate restart cannot create duplicate backend clients
• suppress only intentional transport abort noise and avoid crash/orphan wording for expected child exits
• correct the live and contributor-facing dev_server_status descriptions

Validation

• pnpm run build
• pnpm run test:unit — 233 files, 4292 tests
• pnpm run test:integration — 8 files passed, 2 skipped; 23 tests passed, 5 skipped
• pnpm run lint
• pnpm run check:all-personal-paths
• MCP SDK HTTP smoke, twice: immediate restart, override replace/preserve/clear, status redaction, backend tool forwarding, clean shutdown
• real Docker stdio smoke: repeated restarts and final client shutdown left no ownership-labeled containers
• self-debug smoke through the source dev proxy: breakpoint and locals in the environment merge helper, expression evaluation before/after a step, continue, and output capture

Dogfooding exposed the startup/restart race documented in #541. It was reproduced twice, deduplicated against open and closed issues, filed separately, and fixed in this branch.

Closes #498
Closes #527
Closes #536
Closes #539
Closes #541

[codecov]

Codecov Report

✅ All modified and coverable lines are covered by tests.

📢 Thoughts on this report? Let us know!