[mathiasrw]

Closes #867
Also closes #776, the spike that decided it.

Change

• The default flags of rr and -E runtime are gimu, so the flag u is on: . and [^…] take a whole emoji, [🦖🦕] and 🦖+ work, and \u{1F996} and \p{L} work without a flag.
• --void-unicode (-U) turns it off, like the other --void-… flags, for the old reading as UTF-16 parts.
• -u stays, so old scripts do not fail. With -E regex+ it keeps its meaning: u in place of v.
• --flags still sets all the flags, so --flags gim is the old default. It can not go with -U either.
• When a pattern only fails because of the strict syntax of u, the error says so:

> printf 'a"b' | rexreplace '\"' x
 ❌ Invalid regular expression: invalid escaped character for Unicode pattern
    It works without the flag u, which -U (--void-unicode) turns off.
    No further processing.
> printf 'a"b' | rexreplace '\"' x -U
axb

For the release note of v8: patterns with an escape that has no meaning, like \", \- or \#, a {, } or ] on its own, like a{ or foo], or an octal escape like \01 now stop the run. -U is the way back, and the error says it.

Tests

The tests that checked the old default now say voidUnicode: true, like the compatibility table of #854, which compares rr with the native engine with the same flags. --unicode "is not the default" is now "is the default, and --void-unicode turns it off", and a test checks the hint, and that a pattern that fails also without u gets no hint. The flags in the status report are gimu. The flag combinations of crossover.test.js use voidUnicode. yarn test-ci passes.