Stream zip archive when fetching non-range-request metadata

[charliermarsh]

Summary

If a registry doesn't support range requests, then today, we download the entire wheel to disk and then read the metadata from the downloaded archive. This PR instead modifies the registry client to stream the zipfile and stop as soon as it's seen the metadata, which should be more efficient.

Closes #1596.

Test Plan

Made this the only path for downloading metadata; verified that the test suite passed.

[charliermarsh]

Something like 10-20% faster:

puffin on  charlie/async-meta:main [$!⇡] is 📦 v0.1.6 via 🐍 v3.12.1 (.venv) via 🦀 v1.75.0
❯ python -m scripts.bench \
        --uv-path ./target/release/main \
        --uv-path ./target/release/uv \
        ./scripts/requirements/dtlssocket.in --benchmark resolve-cold --min-runs 50
Benchmark 1: ./target/release/main (resolve-cold)
  Time (mean ± σ):     896.5 ms ±  79.0 ms    [User: 404.9 ms, System: 202.6 ms]
  Range (min … max):   726.3 ms … 1075.0 ms    50 runs

Benchmark 2: ./target/release/uv (resolve-cold)
  Time (mean ± σ):     802.7 ms ±  58.6 ms    [User: 420.7 ms, System: 179.8 ms]
  Range (min … max):   722.5 ms … 953.8 ms    50 runs

Summary
  './target/release/uv (resolve-cold)' ran
    1.12 ± 0.13 times faster than './target/release/main (resolve-cold)'
puffin on  charlie/async-meta:main [$!⇡] is 📦 v0.1.6 via 🐍 v3.12.1 (.venv) via 🦀 v1.75.0 took 1m34s
❯ python -m scripts.bench \
        --uv-path ./target/release/main \
        --uv-path ./target/release/uv \
        ./scripts/requirements/black.in --benchmark resolve-cold --min-runs 50
Benchmark 1: ./target/release/main (resolve-cold)
  Time (mean ± σ):     380.8 ms ±  83.5 ms    [User: 102.9 ms, System: 39.9 ms]
  Range (min … max):   281.8 ms … 550.7 ms    50 runs

Benchmark 2: ./target/release/uv (resolve-cold)
  Time (mean ± σ):     325.3 ms ±  44.7 ms    [User: 113.8 ms, System: 26.9 ms]
  Range (min … max):   269.4 ms … 411.9 ms    50 runs

Summary
  './target/release/uv (resolve-cold)' ran
    1.17 ± 0.30 times faster than './target/release/main (resolve-cold)'