Fix Docker build cache poisoning, add Docker build to PR validation

[AndrewAltimit]

Summary

• Fix Docker build failure: The v0.1.0 release pipeline failed at the Docker image build step because the Dockerfile's dependency pre-caching strategy uses stub source files with --mount=type=cache, but stale cargo fingerprints in the cache prevent cargo from recompiling workspace crates when real source replaces the stubs. This caused 41 compile errors (e.g. could not find 'net' in 'breakpoint_core'). Fixed by purging workspace crate fingerprints from the cache between the stub build and real build.
• Add Docker Build Check to PR validation: The PR pipeline previously had no Docker build step, so this class of failure was invisible until release time. Added a docker-build job that builds the production image and smoke-tests it with a healthcheck, running in parallel with browser tests after CI passes.

Root cause

The Dockerfile uses a two-phase build pattern for dependency caching:

1. Copy Cargo.toml files + create empty stub lib.rs files -> cargo build to compile dependencies into --mount=type=cache
2. COPY crates/ (real source overwrites stubs) -> cargo build again with the same cache mount

The problem: cargo fingerprints in the cache mount persist across steps. When real source replaces stubs, cargo sees matching fingerprints and skips recompiling workspace crates, linking the server binary against the empty stub .rlib files.

Changes

File	Change
docker/server.Dockerfile	Add fingerprint purge step (find ... -name "breakpoint*" -path "*/fingerprint/*" -exec rm) between stub build and real build
.github/workflows/pr-validation.yml	Add docker-build job (build image + healthcheck smoke test), include it in PR status summary

Test plan

• PR validation pipeline runs the new Docker Build Check job successfully
• Re-tag v0.1.0 (or create v0.1.1) after merge to verify release pipeline Docker step passes
• Verify the built Docker image starts and responds to /api/v1/status

Generated with Claude Code