It looks like we can possibly do an out of bounds read if size is equal
to 0.  This commit adds a conditional to ensure size is actually greater
than 0 before looking backwards in the buffer

https://github.com/ruby/prism/commit/2031b626e6

Fixes https://github.com/ruby/prism/pull/2617.

There was an issue with the lexer as follows.
The following are valid regexp options:

```console
$ bundle exec ruby -Ilib -rprism -ve 'p Prism.lex("/x/io").value.map {|token| token[0].type }'
ruby 3.3.0 (2023-12-25 revision https://github.com/ruby/prism/commit/5124f9ac75) [x86_64-darwin22]
[:REGEXP_BEGIN, :STRING_CONTENT, :REGEXP_END, :EOF]
```

The following are invalid regexp options. Unnecessary the `IDENTIFIER` token is appearing:

```console
$ bundle exec ruby -Ilib -rprism -ve 'p Prism.lex("/x/az").value.map {|token| token[0].type }'
ruby 3.3.0 (2023-12-25 revision https://github.com/ruby/prism/commit/5124f9ac75) [x86_64-darwin22]
[:REGEXP_BEGIN, :STRING_CONTENT, :REGEXP_END, :IDENTIFIER, :EOF]
```

As a behavior of Ruby, when given `A` to `Z` and `a` to `z`, they act as invalid regexp options. e.g.,

```console
$ ruby -e '/regexp/az'
-e:1: unknown regexp options - az
/regexp/az
-e: compile error (SyntaxError)
```

Thus, it should probably not be construed as `IDENTIFIER` token.

Therefore, `pm_byte_table` has been adapted to accept those invalid regexp option values.
Whether it is a valid regexp option or not is checked by `pm_regular_expression_flags_create`.
For invalid regexp options, `PM_ERR_REGEXP_UNKNOWN_OPTIONS` is added to diagnostics.

https://github.com/ruby/prism/commit/d2a6096fcf

https://github.com/ruby/prism/commit/bec5ca37a0