Rollup merge of #146699 - heiher:is-ascii-lsx, r=Mark-Simulacrum

Add `is_ascii` function optimized for LoongArch64 for [u8]

Similar to x86_64, on LoongArch64 we use the `vmskltz.b` instruction to test the high bit in a lane.

For longer input cases, the performance improvement is significant. For unaligned cases close to 32 bytes in length, there's some regression, but it seems acceptable.

| core benches (MB/s)                                    | Before | After  | %       |
|--------------------------------------------------------|--------|--------|---------|
| ascii::is_ascii::short::case00_libcore                 | 1000   | 1000   | 0.00    |
| ascii::is_ascii::medium::case00_libcore                | 8000   | 8000   | 0.00    |
| ascii::is_ascii::long::case00_libcore                  | 183947 | 436875 | +137.50 |
| ascii::is_ascii::unaligned_head_medium::case00_libcore | 7750   | 2818   | -63.64  |
| ascii::is_ascii::unaligned_head_long::case00_libcore   | 317681 | 436812 | +37.50  |
| ascii::is_ascii::unaligned_tail_medium::case00_libcore | 7750   | 3444   | -55.56  |
| ascii::is_ascii::unaligned_tail_long::case00_libcore   | 155311 | 436812 | +181.25 |
| ascii::is_ascii::unaligned_both_medium::case00_libcore | 7500   | 3333   | -55.56  |
| ascii::is_ascii::unaligned_both_long::case00_libcore   | 174700 | 436750 | +150.00 |

Author: matthiaskrgr