fix: translate POSIX bracket negation by default

[rajanpanth]

Fixes #187.

[!abc] compiled to a character class containing a literal !, so it matched exactly the characters it is supposed to exclude and rejected everything else:

pm.isMatch('a', '[!abc]')  // true, should be false
pm.isMatch('d', '[!abc]')  // false, should be true
pm.makeRe('[!abc]').source // ^(?:(?:\[!abc\]|[!abc])\/?)$

bash and minimatch both treat [!...] as [^...]. The translation already existed in the tokenizer but was gated behind opts.posix === true, which controls [[:alpha:]] class support; basic bracket negation is POSIX sh glob, not a posix-class feature. This removes the gate, so [!abc] now produces byte-for-byte the same regex as [^abc] and as [!abc] under posix: true.

Behavior notes, all locked by tests:

• [!...] inherits the [^...] slash rule (a negated class never matches /).
• ! not in first position stays literal: [a!] matches a and !.
• Escaped \! in first position stays literal: [\!a] matches ! and a.
• [!] now behaves identically to [^] (both escape the lone ]).

Compatibility: this changes match results for any pattern relying on the inverted behavior. No test in the suite asserted the old behavior (1982 passing after the change, 8 in the new negation block), and the old output contradicts every reference implementation, but flagging it clearly in case you want it to ride a major.

[mrmlnc]

This is the correct change: both Bash and Minimatch treat a leading ! in a bracket expression as negation, so [!abc] is equivalent to [^abc]. There is a nuance with [!], but that behavior predates this PR and should be discussed separately.