[perf] implement Default using const generics

[lcnr]

This PR is mostly just checking if using const generics to implement Default for arrays improves compile times.

cc @MikailBag who is implementing this in #74254

r? @ghost

[lcnr]

@bors try @rust-timer queue

[rust-timer]

Awaiting bors try build completion

[bors]

⌛ Trying commit 329c85d with merge 8e0830658e77454a0c7dd437c75522a6debf9f64...

[bors]

☀️ Try build successful - checks-actions, checks-azure
Build commit: 8e0830658e77454a0c7dd437c75522a6debf9f64 (8e0830658e77454a0c7dd437c75522a6debf9f64)

[rust-timer]

Queued 8e0830658e77454a0c7dd437c75522a6debf9f64 with parent c724b67, future comparison URL.

[rust-timer]

Finished benchmarking try commit (8e0830658e77454a0c7dd437c75522a6debf9f64): comparison url.

[jyn514]

Looks like no (or negligible) effect on perf.

[mati865]

Wall-time, taks-clock and similar time measuring metrics show massive improvement.

Wonder if it's another case for rust-lang/compiler-team#281 or this PR allowed LLVM to better parallelise the work.