Batch prefetch per fork

[konstin]

Previously, the batch prefetcher was part of the solver loop, used across forks. This would lead to each preference in a fork being counted as a tried version, so that after 5 forks with the identical version, we would start batch prefetching. The reported numbers of tried versions are also reported. By tracking the batch prefetcher on the fork the numbers are corrected.

An alternative would be tracking the actually tried versions, but that would mean more overhead in the top level solver loop when the current heuristic works.

In ecosystem/transformers:

$ hyperfine --runs 10 --prepare "rm -f uv.lock" "../../target/release/uv lock --exclude-newer 2024-08-08T00:00:00Z" "uv lock --exclude-newer 2024-08-08T00:00:00Z"
Benchmark 1: ../../target/release/uv lock --exclude-newer 2024-08-08T00:00:00Z
  Time (mean ± σ):     386.2 ms ±   6.1 ms    [User: 396.0 ms, System: 144.5 ms]
  Range (min … max):   378.5 ms … 397.9 ms    10 runs

Benchmark 2: uv lock --exclude-newer 2024-08-08T00:00:00Z
  Time (mean ± σ):     422.0 ms ±   5.5 ms    [User: 459.6 ms, System: 190.3 ms]
  Range (min … max):   415.0 ms … 430.5 ms    10 runs

Summary
  ../../target/release/uv lock --exclude-newer 2024-08-08T00:00:00Z ran
    1.09 ± 0.02 times faster than uv lock --exclude-newer 2024-08-08T00:00:00Z

[charliermarsh]

Nice, thanks -- this makes sense.