New `larry tools cl-format-migrate <dir>`: fixes the "' before prologue start" errors on
Windows-authored tables/xlates whose CRLF line endings choke the Linux Cloverleaf parser
(and Cloverleaf's own hcitblconvert/hcixltconvert). Root cause field-confirmed by Bryan:
`file` shows broken=CRLF/working=LF, both version:6.0 (no format bump) — the stray CR is
read as a bogus char before the prologue keyword.
- Default: strip ONLY trailing CR (sed 's/\r$//'), backup-first via journal.sh, idempotent
(SKIP already-LF), per-file verify (no trailing CRLF remains + prologue block intact),
summary report. Bryan proved a CR-stripped table loads clean.
- Opt-in --run-native-convert: chains hcitblconvert/hcixltconvert after a successful strip
(Cloverleaf's real format bump), resolving the bare name against <site>/Tables|Xlate then
$HCIROOT/Tables|Xlate, one name per call; separate journal checkpoint so a native-step
failure auto-rolls-back to the CR-stripped-safe state. Orphan file -> NATIVE_SKIPPED (never
guesses a site). Default stays strip-only pending Bryan's hcitblconvert-post-strip confirm.
- Unified tbl+xlt (both choke on CRLF identically).
- QoL REPL: cdr->$HCIROOT, cds->$HCIROOT/$HCISITE, cdd->$HCIROOT/$HCISITE/data; bare site name
-> setsite+cd, first-match-wins ordering so it never shadows an existing command.
Vera CODE-PASS (re-gate after fixing a Major verify defect: _count_cr matched any CR byte ->
false FAILED_VERIFY loop on files with a legitimate mid-line CR; now _count_trailing_cr on
\r$, matching the strip). Own-fixture verified: convert/preserve-mid-line-CR/converge-to-SKIP,
full regression + --run-native-convert mocked suite, bash -n 3.2.57, manifest --check clean.
Live-acceptance: Bryan (Gundersen box; dry-run -> single-file --confirm pilot -> batch).
