fix: strip dangling conjunctions from extract_duration remainder

[JarbasAl]

What was broken

extract_duration_generic (the shared engine backing extract_duration for
most languages) consumed each matched <number> <unit> chunk by replacing it
with an empty string. When two chunks in the same sentence were joined by a
conjunction or comma, that connector was left stranded in the remainder text.

>>> extract_duration("It will take about 2 hours and 30 minutes", "en")
(timedelta(seconds=9000), "It will take about and")
>>> extract_duration("2 horas, 30 minutos, y 10 segundos", "es")
(timedelta(seconds=9010), "y")

How it fails

The remainder text is meant to be the sentence with the duration wording
removed, for use by callers (e.g. skills) that want to do something with
"the rest" of the utterance. A dangling "and"/"y" (optionally with commas)
is ungrammatical and can trip up anything downstream that expects clean
leftover text.

The fix

extract_duration_generic now replaces consumed matches with an internal
private-use sentinel character instead of "". Once every unit has been
consumed, a connector (with optional surrounding commas) that is sandwiched
between two sentinels — i.e. it only ever joined two consumed chunks — is
collapsed away along with them. A connector that still touches real,
unconsumed text on either side is left alone, so "two hours and rest and
relax" still returns remainder "and rest and relax".

This is opt-in per language via a new DurationLexicon.connectors field
(a tuple of regex fragments for the language's conjunction word(s)), so it
only changes behavior where explicitly wired up. It's wired up for English
("and") and Spanish ("y"), where the bug was confirmed. Other languages
(e.g. German "und", French "et") keep their existing, already-tested
remainder behavior — those tests intentionally leave the connector in the
remainder and are unaffected.

Verification

Following this repo's NL-TDD convention, failing tests were written first
to reproduce the bug (test/parse_tests/test_parse_en.py and
test/parse_tests/test_parse_es.py), then the fix was applied and the
tests turned green.

Ran the full suite before and after the change (PYTEST_DISABLE_PLUGIN_AUTOLOAD=1 pytest test/): the same 14 pre-existing failures are present on dev
before this change (unrelated test_dates_an* and test_packaging
failures) — zero new failures, zero regressions.

---

Authored with AI assistance (Claude).

[coderabbitai]

Important

Review skipped

Draft detected.

Please check the settings in the CodeRabbit UI or the .coderabbit.yaml file in this repository. To trigger a single review, invoke the @coderabbitai review command.

⚙️ Run configuration

Configuration used: defaults

Review profile: CHILL

Plan: Pro Plus

Run ID: 5887e111-030f-4e59-81c5-292877d37ed6

You can disable this status message by setting the reviews.review_status to false in the CodeRabbit configuration file.

Use the checkbox below for a quick retry:

• 🔍 Trigger review

✨ Finishing Touches

🧪 Generate unit tests (beta)

• Create PR with unit tests
• Commit unit tests in branch fix/duration-remainder-conjunctions

---

Thanks for using CodeRabbit! It's free for OSS, and your support helps us grow. If you like it, consider giving us a shout-out.

❤️ Share

• X
• Mastodon
• Reddit
• LinkedIn

_{Comment @coderabbitai help to get the list of available commands.}