feat(regexp): regular expressions library (Go RE2)

[jig]

Summary

Adds regular expressions to jig/lisp — there was none. A regexp namespace, loaded by default in the lisp binary, with Clojure-style semantics on Go's RE2 engine.

(re-matches? ¬\d+¬ "123")                 ; => true   (anchored — whole string)
(re-find?    ¬\d+¬ "abc123")              ; => true   (substring)
(re-find ¬(\d+)-(\d+)¬ "call 555-1234")   ; => ["555-1234" "555" "1234"]
(get (re-find ¬v(\d+)¬ "v42") 1)          ; => "42"

• re-pattern compiles a first-class, reusable regex (prints «regex …»). Every function also accepts a raw pattern string (compiled on use), so re-pattern is optional.
• re-matches? / re-find? → booleans (anchored whole-string / unanchored substring) — the primary need.
• re-matches / re-find → Clojure-style match data: nil, the match string when there are no groups, or a vector [whole g1 g2 …] (an unmatched optional group is nil, faithful to Clojure via FindStringSubmatchIndex).

Design notes

• RE2, not PCRE: no backreferences or lookaround; linear-time in exchange. Documented.
• No #"…" reader literal: jig/lisp's raw ¬…¬ string already avoids escaping (¬\d+¬; a normal "\d+" is a read error), so it's the natural pattern syntax — no scanner changes.
• re-matches anchors with \A(?:…)\z so "whole string" holds regardless of a (?m) flag.
• re-replace / re-split / re-seq are deferred to a later iteration (agreed scope).

Test plan

• lib/regexp/regexp_test.go: predicates (anchored vs substring, compiled vs raw), match data (no-groups → string, groups → vector, unmatched group → nil, no match → nil), the pattern value («regex …» print, re-pattern idempotent), and errors (bad pattern, wrong arg types).
• Registered in main.go and docgen/libraries.go (sync test passes); LANGUAGE.md regenerated. Full suite green with and without -tags debugger.

🤖 Generated with Claude Code