fix: escape DEL, C1 controls, and invalid UTF-8 in log output

[tobez]

log_enc() escaped C0 controls, quotes, and backslashes, but passed bytes
0x7F–0xFF through raw — in bare tokens as well as quoted values. DEL and
single-byte C1 controls (e.g. 0x9B CSI) could therefore reach the terminal
unescaped, and malformed byte sequences could corrupt log-line structure.

The encoder is now UTF-8 aware, in the spirit of Go's log/slog:

• Non-ASCII input is validated as strict UTF-8 (RFC 3629): overlong
  encodings, surrogates, and code points above U+10FFFF are rejected.
• Valid sequences encoding U+00A0 and above pass through unchanged, so
  legitimate non-ASCII text is unaffected.
• DEL (0x7F), C1 controls (U+0080–U+009F), and invalid UTF-8 are escaped
  byte-wise as \xNN.
• needs_quote() applies the same classification, so bare tokens
  containing such bytes are now quoted and escaped.

Test plan

• 18 new log_enc() cases in t/test_log.c (DEL, C1 raw and encoded,
  overlongs, surrogates, out-of-range, truncated sequences, resync after a
  bad lead byte, and pass-through of valid 2/3/4-byte UTF-8)
• full suite passes (367 tests)
• scan-build: no bugs found