Parse all lines, extracting timestamps where possible.
Lines that don’t match any timestamp pattern are included as plain-text entries.
Caches the detected format from the first match and tries it first on subsequent
lines, avoiding up to 3 wasted regex attempts per line.