ci: add transport performance regression detection

[iduartgomez]

Summary

Add CI workflow that automatically detects performance regressions in the transport layer by running level0 benchmarks on PRs.

How it works

1. When main branch gets a push: Runs benchmarks and saves baseline to cache
2. When PR is opened/updated: Restores main's baseline, runs benchmarks, compares
3. If regression detected: Posts comment on PR with details

What it tests (level0 benchmarks)

Benchmark	What it measures
level0/crypto/encrypt	AES-GCM encryption throughput
level0/crypto/decrypt	AES-GCM decryption throughput
level0/crypto/nonce	Nonce generation speed
level0/serialization	Packet serialization overhead
level0/packet_creation	Full packet creation pipeline
level0/memcpy	Memory copy baseline

These are pure logic benchmarks - deterministic, no I/O, no async runtime overhead.

Key design decisions

• Non-blocking: PRs can still merge even if regressions detected (informational only)
• Criterion comparison: Uses Criterion's built-in baseline comparison with 10% noise threshold
• Cache strategy: Each main commit creates a new baseline; PRs use restore-keys to find most recent
• Self-hosted runner: Runs on self-hosted for consistent hardware

Test plan

• Merge this PR to establish initial baseline
• Open a test PR touching transport code
• Verify workflow runs and compares against baseline
• Verify PR comment appears if regression introduced

🤖 Generated with Claude Code