Fix CI/CD Pipeline Inconsistencies (#13)

* fix(ci): resolve CI/CD inconsistencies and broken baseline storage

This commit addresses several critical issues in the CI/CD workflows:

1. Fix broken baseline artifact storage:
   - benchmark.yml now downloads baseline from update-baseline.yml workflow
   - Previously tried to download from itself after baseline upload was removed

2. Standardize Rust toolchain versions to stable:
   - CI workflow: Changed test and rustfmt jobs from nightly to stable
   - CD workflow: Changed deb publish job from nightly to stable
   - Ensures consistent builds and testing across all jobs

These changes restore benchmark comparisons and ensure consistent CI/CD behavior.

* feat(ci): add /bench command for on-demand PR benchmark comparisons

Add new workflow that allows repository owners to trigger benchmark
comparisons between any two refs via PR comments.

Features:
- Command syntax: /bench <ref1> <ref2>
- Owner-only security: Only repo owner can trigger
- Works on PR comments only (not regular issues)
- Compares any two commits, branches, or tags
- Posts detailed comparison report to PR
- Interactive reactions (👀 → 🚀 or 😕)

Use cases:
- Compare feature branch vs stable release
- Validate optimizations between specific commits
- Test performance across version boundaries
- Ad-hoc performance debugging

Documentation:
- Updated scripts/README.md with complete workflow documentation
- Added security model and usage examples
- Documented all three benchmark workflows

* refactor(ci): simplify benchmark system to single /bench command

Removed automatic benchmarking workflows and consolidated to a single
on-demand command-based system with enhanced flexibility.

Changes:
- Removed benchmark.yml (automatic benchmarks on every PR/push)
- Removed update-baseline.yml (manual baseline updates)
- Enhanced bench-command.yml with optional parameters:
  * /bench <ref1> <ref2> [iterations] [sizes]
  * Defaults: iterations=100, sizes=1000,5000,10000
  * Examples: /bench main HEAD
             /bench v0.12.0 v0.13.0 100 1000,5000,10000

New features:
- Graceful handling when benchmark tool doesn't exist in refs
- Posts helpful error messages with commit info
- Validates all parameters before running
- Supports custom iterations and sizes per comparison
- Build logs included in artifacts for debugging

Benefits:
- No automatic runs = lower CI costs
- Owner controls when benchmarks run
- Flexible parameters per comparison
- Clear error messages for invalid refs
- Only one workflow to maintain

Documentation:
- Completely rewrote scripts/README.md
- Added comprehensive usage examples
- Documented all error cases
- Added use case scenarios
- Simplified troubleshooting guide

* fix(ci): improve bench workflow robustness and reliability

Address critical issues to make the workflow less flaky and more reliable.

Fixes:

1. Permissions
   - Added issues: write and pull-requests: write to check-permission job
   - Allows posting comments and reactions without permission errors

2. Concurrency control
   - Added concurrency group: bench-${{ issue.number }}
   - Prevents overlapping benchmark runs on the same PR
   - Uses cancel-in-progress: false to queue instead of cancel

3. Ref fetching
   - Added targeted git fetch before validation
   - Fetches refs, tags, and heads explicitly
   - Prevents "ref not found" errors for remote branches/tags

4. Shell robustness
   - Added set -euo pipefail to all bash scripts
   - Ensures early failure on any error
   - Prevents silent failures and undefined variables

5. Early parse feedback
   - Added continue-on-error to parse step
   - Posts immediate error message on invalid format
   - Shows usage examples with all parameters
   - Prevents workflow from proceeding with bad params

Benefits:
- More reliable ref resolution
- Clear error messages for invalid input
- No concurrent runs causing conflicts
- Proper permissions for all operations
- Fail-fast behavior prevents wasted CI time

* fix(ci): fix YAML syntax errors and improve workflow behavior

Critical fixes for YAML parsing and workflow reliability:

1. YAML Syntax Errors (CRITICAL)
   - Fixed template literals with markdown causing YAML parse errors
   - Converted all multi-line template literals to array.join() pattern
   - Lines starting with ** were interpreted as YAML alias markers
   - Affected: parse error message, acknowledgment, results, error handler

2. Graceful Error Handling
   - Removed process.exit(78) from "Post missing tool error" step
   - Non-zero exit triggers handle-error job unnecessarily
   - Existing if guards properly skip subsequent steps
   - Workflow now exits gracefully without false failures

3. Concurrency Behavior
   - Changed cancel-in-progress from false to true
   - Previous: Multiple /bench commands queue and run sequentially
   - Now: Newer /bench commands cancel older pending runs
   - Saves CI time and provides faster feedback

All changes maintain functionality while fixing syntax and improving UX.

* feat(ci): add intelligent commit ordering and same-commit detection

Implement the intended benchmark logic to ensure correct comparisons:

1. Automatic commit ordering
   - Determines which ref is older (baseline) vs newer (current)
   - Uses git commit timestamps for ordering
   - Ensures baseline is always the older commit regardless of argument order
   - Example: /bench v0.13.0 main → baseline=v0.13.0, current=main (if main is newer)

2. Same-commit detection
   - Checks if both refs resolve to the same SHA
   - Exits early with helpful message if refs are identical
   - Prevents wasteful benchmark runs on identical code
   - Example: /bench main origin/main → detects same commit, posts message

3. Updated all steps to use determined ordering
   - Renamed steps: check_ref1_tool → check_baseline_tool, check_ref2_tool → check_current_tool
   - Updated benchmark steps to use baseline_ref and current_ref
   - Updated artifact names: benchmark_ref1.json → benchmark_baseline.json
   - Updated build logs: build_ref1.log → build_baseline.log
   - Results always show: Baseline (older) vs Current (newer)

4. Improved error messages
   - Missing tool errors now indicate "baseline (older)" vs "current (newer)"
   - Same-commit message explains refs are identical
   - Clear distinction between user input and determined ordering

Benefits:
- Consistent baseline/current semantics regardless of argument order
- No wasted CI time on identical commits
- Clear, predictable results in every scenario
- Better user experience with informative messages

---------

Co-authored-by: Claude <noreply@anthropic.com>

Author: lalvarezt