Update the current replace implementation for both vim and helix to take
into account the grapheme count instead of the unicode scalar count.

Using the unicode scalar count was already an improvement compared to
what we had before, but it doesn't cover all scenarios, for example,
when using some emoji, which although unlikely in source code, is a
possibility and is currently correctly handled by Neovim.

This commit introduces a new
`multi_buffer::MultiBufferSnapshot::grapheme_count_for_range` method
which returns the number of graphemes for a given range in the
multibuffer, which then allows us to correctly determine the number of
times the character being used for replace should repeat.

Existing tests have also been updated with test cases that would fail
without these new changes, to catch any potential regressions.