mut and par-each optimizations for performance

[fdncred]

Description

This PR has a few changed to increase performance with par-each and mut assignments.

Two performance optimizations targeting common hot paths in Nushell: mut field assignment (cloning reduction via get_var_mut) and par-each (thread-pool reuse + captures-only stack clone).

I also added a few benchmarks.

User-facing changes (Release notes)

• Improved performance of mut variable field assignment (e.g., $list.field = value, $table.0 = value): the variable is now mutated in-place on the evaluation stack instead of being cloned, modified, and stored back. Large list variables see the most benefit — up to 70× faster for 100k-element lists.

• Improved performance of repeated par-each calls: custom thread pools (--threads N) are cached by thread count and reused, eliminating OS thread spawn overhead on repeated calls. The streaming (unordered) path now only clones the closure's captured variables instead of the entire stack. Combined, these reduce par-each per-call overhead by up to 33% when calling par-each many times sequentially.

Additional notes

[Tyarel8]

Using latest nightly with a command like glob "~/scoop/apps/{uv}/**/*.exe" | wrap name | par-each {} (I have remove the code inside the par-each since the bug happens even if it's empty), it hangs almost always, I haven't bisected the issue, but I'd guess this pr is the cause.

[fdncred]

i'll look into it, thanks.

[fdncred]

@Tyarel8 can you try out #18628 ?

[Tyarel8]

Works for me👍