ci: stage push benchmarks + label-driven auto-tag releases

[EshwarCVS]

What this adds

• Run benchmark workflow on direct stage pushes (in addition to PRs).
• Add auto-tag-release.yml: when a PR is merged into master and has one of these labels:
  • release:patch
  • release:minor
  • release:major
    it creates and pushes the next vX.Y.Z tag automatically.

Why

• stage gets stronger continuous signal before promotion.
• master releases remain controlled (label-gated), but no manual tag command is needed.
• Existing tag-triggered release workflow continues to publish to PyPI/Docker/GitHub Release.

Note

• No label => no release tag.

[github-actions]

Benchmark results

Ubuntu runner, Python 3.13. HTTP table uses the same httpx load against uvicorn (Python) and Fiber (Go). Direct ASGI (below) is Python-only and excludes network I/O.

HTTP throughput (FasterAPI vs FastAPI vs Fiber)

Endpoint	FasterAPI	FastAPI	Fiber (Go)	F / Fast
GET /health	481/s	506/s	489/s	0.95x
GET /users/{id}	531/s	548/s	539/s	0.97x
POST /users	481/s	488/s	480/s	0.99x

Direct ASGI (no HTTP; 50,000 iterations)

Endpoint	FasterAPI	FastAPI	Speedup	vs README ASGI ratio
GET /health	197,652/s	19,516/s	10.13x	+47.9%
GET /users/{id}	183,308/s	13,740/s	13.34x	+52.8%
POST /users	108,236/s	11,111/s	9.74x	+36.2%

Routing (radix vs regex, 1,500,000 lookups)

Router	Ops/s	Speedup	vs README
Radix	954,440	9.7x	+27.1%
Regex	98,842	1.0x

How to read this

• F / Fast = FasterAPI req/s ÷ FastAPI req/s on the same HTTP harness (higher is better).
• Fiber uses the Go app in benchmarks/fiber (same routes). Go is often several times faster than Python here; the important guard for regressions is check_regressions.py (ASGI + routing floors), which must pass in this workflow.
• vs README compares combined speedups to documented reference numbers (local machine); CI absolute req/s differs by hardware.