Poor codegen for baseline x86_64 std::to_chars(base=2)

As of @philnik777's #210659, libc++ vectorizes std::to_chars(base=2) (and 16).

However, codegen is quite poor on baseline x86_64 (without avx512), emitting 38 instructions. I can do it in 23. https://godbolt.org/z/na33fhdzz

(For avx512, Clang emits 14 instructions, but I can do 12 https://godbolt.org/z/qjzcnW8q9)

Somewhat discussed on ms-stl discord https://discord.com/channels/737189251069771789/922607955621318697/1528730955823710419

Start with the std::to_chars(base=2) codegen shown in the linked Godbolt examples, comparing baseline x86_64 without AVX-512 against the AVX-512 result. Identify where the emitted instruction counts differ from the examples, then verify that the baseline and AVX-512 codegen improve without regressing the existing behavior.

Written by the indexing model from the issue text.