Add web framework benchmark example (Barq)

[grandimam]

Demonstrates free-threaded Python performance benefits for web frameworks:

• Pure Python HTTP framework using ThreadPoolExecutor
• 2-5x faster than FastAPI on various workloads
• Shows thread scaling with CPU-bound handlers

[ngoldbaum]

@nascheme would you mind giving this a look, along with barq? You know a lot more about web development than I do.

[nascheme]

I think this is a pretty good example of how to use free-threaded Python to build such a thing. The HTTP/1.1 protocol implementation is skeletal, let's say. The readme has an "Experimental" note and that's good. As an example for people to learn from, I think the simpler code is an advantage.

I have submitted a PR that updates it to use Python 3.14t. I think it would be a good idea even though I'm not sure how much extra performance that gives. 3.14 has been released for a while now so there should be no problem in using it.

[nascheme]

Oh, one additional suggestion for @grandimam, I think this statement in your readme might be misleading:

2-5x faster than FastAPI on real workloads.

How have you tested it on real workloads? If you are talking about the scripts inside benchmarks then I think it's quite a stretch to call those "real". I'd suggest re-wording that. In this PR, you say "various workloads".

[nascheme]

For interests sake, here are benchmark results on my 28-core i7-14700K. I'm using 10 barq workers.

The difference between 3.13t and 3.14t seems primarily noise, would need more complex benchmark tests to see if one is actually faster.

============================================================
  BARQ vs FASTAPI (optimal configs)
  2000 requests, 100 concurrent clients
============================================================
  Barq: 10 threads, blocking I/O
  FastAPI: async + aiosqlite
============================================================

Starting servers...
Servers ready.

─── JSON ───
  Barq (10 threads)
    2000/2000 ok, 12002.0 req/s
    latency: avg=2.1ms p50=1.9ms p99=7.2ms
  FastAPI (async)
    2000/2000 ok, 4975.2 req/s
    latency: avg=19.3ms p50=18.6ms p99=28.5ms
    → Barq 141.2% faster

─── DB ───
  Barq (10 threads)
    2000/2000 ok, 40472.5 req/s
    latency: avg=2.1ms p50=2.0ms p99=4.4ms
  FastAPI (async)
    2000/2000 ok, 2587.5 req/s
    latency: avg=38.2ms p50=34.7ms p99=64.6ms
    → Barq 1464.1% faster

─── CPU ───
  Barq (10 threads)
    2000/2000 ok, 3341.4 req/s
    latency: avg=29.1ms p50=29.5ms p99=32.1ms
  FastAPI (async)
    2000/2000 ok, 368.8 req/s
    latency: avg=264.5ms p50=270.0ms p99=278.1ms
    → Barq 806.0% faster