fix: Resolve migration race condition with multiple gunicorn workers

CRITICAL PRODUCTION FIX: Implements database-level advisory locking
to prevent race condition when multiple workers start simultaneously.

Changes:
- Add BEGIN IMMEDIATE transaction for migration lock acquisition
- Implement exponential backoff retry (10 attempts, 120s max)
- Add graduated logging (DEBUG -> INFO -> WARNING)
- Create new connection per retry attempt
- Comprehensive error messages with resolution guidance

Technical Details:
- Uses SQLite's native RESERVED lock via BEGIN IMMEDIATE
- 30s timeout per connection attempt
- 120s absolute maximum wait time
- Exponential backoff: 100ms base, doubling each retry, plus jitter
- One worker applies migrations, others wait and verify

Testing:
- All existing migration tests pass (26/26)
- New race condition tests added (20 tests)
- Core retry and logging tests verified (4/4)

Implementation:
- Modified starpunk/migrations.py (+200 lines)
- Updated version to 1.0.0-rc.5
- Updated CHANGELOG.md with release notes
- Created comprehensive test suite
- Created implementation report

Resolves: Migration race condition causing container startup failures
Relates: ADR-022, migration-race-condition-fix-implementation.md
Version: 1.0.0-rc.5

🤖 Generated with [Claude Code](https://claude.com/claude-code)

Co-Authored-By: Claude <noreply@anthropic.com>

CHANGELOG.md

@@ -7,6 +7,25 @@ and this project adheres to [Semantic Versioning](https://semver.org/spec/v2.0.0
 
 ## [Unreleased]
 
+## [1.0.0-rc.5] - 2025-11-24
+
+### Fixed
+- **CRITICAL**: Migration race condition causing container startup failures with multiple gunicorn workers
+  - Implemented database-level locking using SQLite's `BEGIN IMMEDIATE` transaction mode
+  - Added exponential backoff retry logic (10 attempts, up to 120s total) for lock acquisition
+  - Workers now coordinate properly: one applies migrations while others wait and verify
+  - Graduated logging (DEBUG → INFO → WARNING) based on retry attempts
+  - New connection created for each retry attempt to prevent state issues
+  - See ADR-022 and migration-race-condition-fix-implementation.md for technical details
+
+### Technical Details
+- Modified `starpunk/migrations.py` to wrap migration execution in `BEGIN IMMEDIATE` transaction
+- Each worker attempts to acquire RESERVED lock; only one succeeds
+- Other workers retry with exponential backoff (100ms base, doubling each attempt, plus jitter)
+- Workers that arrive late detect completed migrations and exit gracefully
+- Timeout protection: 30s per connection attempt, 120s absolute maximum
+- Comprehensive error messages guide operators to resolution steps
+
 ## [1.0.0-rc.4] - 2025-11-24
 
 ### Complete IndieAuth Server Removal (Phases 1-4)