feat(vendor): support modifying remote files in vendor folder without checksum errors

[dsherret]

Includes:

• feat: use loader to verify https checksums deno_graph#486
• perf: move jsr manifest checksum to locker deno_graph#488
• feat: remove checking checksums from lockfile crate deno_lockfile#25
• refactor(BREAKING): move file system functionality out of lockfile crate deno_lockfile#22
• fix: allow invalid utf8 when decoding modules deno_graph#483
• fix: handle return statements in functions correctly deno_graph#470

This is actually a lockfile perf investigation of moving the lockfile verification to be done in deno_graph's loader, but it has the side effect of enabling the ability to modify remote modules in the vendor folder without causing checksum errors.

Benchmarks

% git clone https://github.com/jsr-io/jsr
% git switch deno_graph_slow
% cd frontend
% hyperfine --warmup 3 "../../deno/deno_old run -A repro.ts" "../../deno/target/release/deno run -A repro.ts"

With "vendor": true (note: this PR doesn't fix the underlying issue that makes this scenario slow... I will follow up with that later):

Benchmark 1: ../../deno/deno_old run -A repro.ts
  Time (mean ± σ):     538.0 ms ±   5.5 ms    [User: 444.0 ms, System: 101.5 ms]
  Range (min … max):   531.9 ms … 548.6 ms    10 runs
 
Benchmark 2: ../../deno/target/release/deno run -A repro.ts
  Time (mean ± σ):     458.5 ms ±   3.1 ms    [User: 366.4 ms, System: 100.5 ms]
  Range (min … max):   453.4 ms … 463.7 ms    10 runs
 
Summary
  ../../deno/target/release/deno run -A repro.ts ran
    1.17 ± 0.01 times faster than ../../deno/deno_old run -A repro.ts

With "vendor": false (slightly slower because it now has to verify checksums):

Benchmark 1: ../../deno/deno_old run -A repro.ts
  Time (mean ± σ):     539.5 ms ±   3.7 ms    [User: 447.8 ms, System: 99.8 ms]
  Range (min … max):   536.2 ms … 546.0 ms    10 runs
 
Benchmark 2: ../../deno/target/release/deno run -A repro.ts
  Time (mean ± σ):     463.6 ms ±   8.3 ms    [User: 371.3 ms, System: 100.8 ms]
  Range (min … max):   456.0 ms … 484.2 ms    10 runs

[dsherret]

Fixed!

[dsherret]

Handled in graph_roots_valid.

[dsherret]

It seems some wpt tests are failing due to denoland/deno_graph#483

See https://github.com/denoland/deno/actions/runs/9262872207

I'm unsure how to interpret the wpt test output. Are we incorrectly failing now?

[lucacasonato]

I have updated the expectations. We were previously failing these tests, now the tests pass. The expectations file was assuming the tests to fail though, so them passing caused the suite to fail.

[dsherret]

Thanks!

[marvinhagemeister]

LGTM, went through the PR 👍