Loosen regex pin to allow calendar versions past 2024

[eljero66]

Problem

regex uses date-based (calendar) versioning, not semver. The current constraint

regex = "^2024.11.6"

expands (via Poetry's caret rule) to >=2024.11.6,<2025.0.0, which locks out every 2025+ release of regex — treating the calendar year 2024 as if it were a semver major. Those newer releases are not breaking changes; they're routine bugfix/feature releases under the library's date-tag scheme.

This blocks any environment that needs a newer regex alongside deduce — for example co-installing transformers>=5.5.0, which requires regex>=2025. Today the two are mutually exclusive purely because of this caret.

Why it's safe

deduce uses the regex library in exactly one place:

# deduce/tokenizer.py
_TOKENIZER_PATTERN = regex.compile(r"\w+|[\n\r\t]| {4,}|[^ ]", flags=regex.I | regex.M)

That's a single regex.compile using \w+ and the I/M flags — bedrock API that has been stable for years. (The BSN and phone annotators use the stdlib re, not this library.)

Change

Turn the cap into a floor:

-regex = "^2024.11.6"
+regex = ">=2024.11.6"

poetry.lock regenerated to match (still resolves regex to 2024.11.6 by default — the change only permits newer versions, it doesn't force them).

Verification

With regex 2026.9.3 installed (≈2 years past the old cap):

• Full test suite: 122 passed, 0 failures.
• End-to-end smoke test unaffected:

  in : Jan de Vries woont in Amsterdam en belt vaak naar 0612345678.
  out: [PERSOON-1] woont in [LOCATIE-1] en belt vaak naar [ID-1].
  

Happy to adjust the constraint form (e.g. an explicit upper bound at a future known-good version) if you prefer a different policy for this dep.