[PERF][WIP] Use unchecked_add in Layout::repeat

[ecstatic-morse]

#69635 (a rollup) caused a small performance regression. Nothing in that rollup jumps out at me, but #69544 alters some code that is pretty hot, so let's mess with that.

r? @ghost (just for a perf run)

[ecstatic-morse]

@bors try
@rust-timer queue

[rust-timer]

Awaiting bors try build completion

[bors]

⌛ Trying commit 94045da with merge 9cea05f9131a9149b2683c42b30e78ccf4c4f45d...

[jonas-schievink]

#69544 changes this from checked_add to +, which I'd expect to improve performance. Very odd. I also can't see anything else in that rollup though.

[bors]

☀️ Try build successful - checks-azure
Build commit: 9cea05f9131a9149b2683c42b30e78ccf4c4f45d (9cea05f9131a9149b2683c42b30e78ccf4c4f45d)

[rust-timer]

Queued 9cea05f9131a9149b2683c42b30e78ccf4c4f45d with parent 3c5b1b7, future comparison URL.

[rust-timer]

Finished benchmarking try commit 9cea05f9131a9149b2683c42b30e78ccf4c4f45d, comparison URL.

[ecstatic-morse]

No change. We could try reverting #69544 entirely or adding #[inline(never)] to repeat, but the perf queue is pretty full so I don't want to do too many speculative perf runs. I think I'll close this and open an issue.