chore(turbopack): Update dashmap from 5.x to 6.x (#72433)
Dashmap 6.x claims fairly large performance improvements:
> This release contains performance optimizations, most notably 10-40% gains on Apple Silicon but also 5-10% gains when measured in Intel Sapphire Rapids.
-- https://github.com/xacrimon/dashmap/releases/tag/v6.0.0
I don't think this means too much for us because while `turbo-tasks` uses dashmap pretty heavily, we spend more time actually doing computations than in `turbo-tasks`, so it's a fraction of a fraction.
Still, here's a fairly unscientific (noisy environment, inside a VM) benchmark on my macbook using a microbenchmark that's designed to hit `turbo-tasks` extremely hard (much more so than would happen in realistic scenarios):
```
$ TURBOPACK_BENCH_STRESS=yes cargo bench -p turbo-tasks-memory -- fibonacci/200
Compiling turbo-tasks-memory v0.1.0 (/home/bgw.linux/next.js/turbopack/crates/turbo-tasks-memory)
Finished `bench` profile [optimized] target(s) in 2.80s
Running benches/mod.rs (target/release/deps/mod-7b9a423ac1e10dce)
Gnuplot not found, using plotters backend
Benchmarking turbo_tasks_memory_stress/fibonacci/200: Warming up for 3.0000 s
Warning: Unable to complete 2000 samples in 5.0s. You may wish to increase target time to 91.9s, or reduce sample count to 100.
turbo_tasks_memory_stress/fibonacci/200
time: [33.008 ms 33.108 ms 33.208 ms]
thrpt: [605.31 Kelem/s 607.13 Kelem/s 608.97 Kelem/s]
change:
time: [-2.0573% -1.6309% -1.2221%] (p = 0.00 < 0.05)
thrpt: [+1.2373% +1.6580% +2.1005%]
Performance has improved.
Found 44 outliers among 2000 measurements (2.20%)
2 (0.10%) low mild
38 (1.90%) high mild
4 (0.20%) high severe
```
Any potential improvement to realistic topline numbers are going to be <1% and thus too small to measure (especially since I don't have a low-noise way of measuring Apple Silicon where most of the benefits supposedly are, only x86-64).
Closes PACK-3411
