Fix more ancient bugs around Latin-1 handling.

It turned out that case folding assumed UTF-8 mode, so
we would fold, say, 0xD1 to 0xF1 even in Latin-1 mode.

Fixes #477.

Change-Id: I73aa5c8e33ee0c6041c54e3a7268635915960f64
Reviewed-on: https://code-review.googlesource.com/c/re2/+/62714
Reviewed-by: Alex Chernyakhovsky <achernya@google.com>
Reviewed-by: Paul Wankadia <junyer@google.com>