Fix race condition in parallel FastGA execution

This commit resolves intermittent segfaults when multiple FastGA
processes run concurrently by implementing two key fixes:

FIX 1: Use unique temp directories per FastGA invocation
- Create a unique temp directory for each FastGA process
- Pass it via the -P flag to prevent collisions in shared /tmp
- Prevents race conditions in GDB file creation and temp file handling

FIX 2: Use tempfile crate for .1aln output
- Replace PID-based naming with tempfile::Builder for guaranteed uniqueness
- Provides atomic file creation even under heavy contention
- Defense-in-depth against temp file collisions

Testing:
- All integration tests pass with --test-threads=8 (previously failing at 30-50% rate)
- Verified stability across multiple test runs
- No performance regression observed

CI Changes:
- Restored parallel test execution (removed --test-threads=1 workaround)
- Expected to reduce CI time from 12min back to ~5min

References: See RACE_CONDITION_ANALYSIS.md for detailed investigation

Author: ekg

.github/workflows/ci.yml

@@ -36,7 +36,7 @@ jobs:
         run: cargo build --release
 
       - name: Run tests
-        run: cargo test --release -- --test-threads=1
+        run: cargo test --release
 
       - name: Run clippy
         run: cargo clippy -- -D warnings