feat(core): "digits" mode for spoken digit-by-digit sequences

[jqueguiner]

The cardinal grammar mis-parses digit-by-digit spellings — "two seven five" → 14, "seven eight three one eight" → 27 — but dictated house numbers, zip codes and phone-style runs are commonly spelled this way (found analysing ASR transcripts of spoken postal addresses).

Fix

New Converter::to_digits + "digits" conversion type: each whitespace token of a run must be a single-digit number word (0-9) in the target language (via the existing per-language cardinal converter); the run is concatenated instead of summed. Any non-digit or 10+ token makes it decline, so the sentence walker falls back to its longest cardinal parse — a genuine cardinal is never turned into a digit string.

words2num_sentence("Two seven five", "en", "digits")            -> "275"
words2num_sentence("Seven Eight Three One Eight Ave", "en", ..) -> "78318 Ave"
words2num_sentence("one eight nine zero zero one one", "en", ..)-> "1890011"
words2num_sentence("sieben acht drei", "de", "digits")          -> "783"
words2num_sentence("sixty-one hundred Main St", "en", "digits") -> unchanged

• Language-agnostic (reuses the reverse tables).
• Non-breaking (new opt-in to type; passes through the Python binder unchanged).
• Unit test added; digits added to the CLI help.
• Limitation: the numeric value type collapses a leading zero ("zero one" → 1).

[jqueguiner]

Superseded by a clean-base branch to clear a stale GitHub merge-base glitch.