Fix bug in GET_PREV_CHAR macro

[chqrlie]

• pass cbuf_type variable to XXX_CHAR macros in lre_exec_backtrack()
• improve readability of these macros
• fix GET_PREV_CHAR macro: cptr was decremented twice on invalid high surrogate.

[chqrlie]

I found this while proof reading as I was merging some of the differences into quickjs. It is difficult to create a test case triggering this bug...

[bnoordhuis]

Can you leave out stylistic changes? They make the diff larger and more difficult to read.

As a reviewer, I first have to figure out if it's actually a functional change - and if not, if it's really equivalent to the old code. As a maintainer, style changes introduce churn that makes git blame less effective. As a contributor, it's rage inducing when your carefully crafted pull request ends up with a ton of merge conflicts because someone landed a style change the other day.

Changes that improve legibility are okay but a) sparingly, and b) should be split off into separate commits and marked NFC (Non-Functional Change.)

[chqrlie]

Sorry about that, I was trying to reduce the macro size and combine the tests for consistency. I posted a patch with minimal NFC differences.

[chqrlie]

I am going to merge this patch. This is work in progress:

• we might use unions for the cbuf_xxx pointers to avoid warnings about alignment in casts
• there is substantial room for performance improvement in lre_exec_backtrack handling some opcodes more explicitly for the 3 different supported encodings.
• we should have variable length opcodes for string fragment matches, or at least fixes length string fragments.
• we could also use different opcodes to avoid testing s->ignore_case and s->unicode for each character.
• these changes would pave the way to generate JIT code for regexp matching.