fix: Keep a zero watermark delay threshold on a Kafka source

[Rodrigo-Palma]

KafkaOptions.__init__ stored watermark_delay_threshold or None (data_source.py:50), and timedelta(0) is falsy, so a delay of zero was read as "not given":

from datetime import timedelta
from feast.data_source import KafkaOptions

o = KafkaOptions("localhost:9092", fmt, "t", watermark_delay_threshold=timedelta(0))
o.watermark_delay_threshold                             # None
o.to_proto().HasField("watermark_delay_threshold")      # False

The value never reached the proto, so it was lost on apply() and on every registry round-trip, while timedelta(seconds=30) came back intact. Zero is a meaningful setting for a stream source: it asks for no watermark delay at all, which is not the same as leaving the option out.

Two things were keeping this hidden.

The branch that was meant to preserve a zero could not return one. Both from_proto sites mapped an explicit zero Duration back to timedelta(days=0), and then handed it to the constructor above, which dropped it. Feeding KafkaOptions.from_proto a proto that does carry a zero Duration returned None.

KafkaSource.from_proto tested the message for truth instead of for presence (data_source.py:555). A protobuf message object is always truthy, so an absent delay entered the branch too and came out as timedelta(days=0); only the or None in the constructor hid that. Fixing the constructor alone would have turned this into a visible regression, which is why both are in this PR. KafkaOptions.from_proto, its sibling in the same file, already reads the field with HasField, and both now do.

After the change: an absent delay stays None and is not written to the proto, a zero delay stays timedelta(0) and survives a round-trip, and a non-zero delay is unchanged.

Test run: sdk/python/tests/unit/test_data_sources.py, 14 passed; the new test fails on master with assert None == datetime.timedelta(0). ruff check and ruff format --check clean on both files with the pinned 0.16.4. (test_stream_feature_view.py fails 3 tests here for a missing pyspark, the same way on master.)

[codecov-commenter]

⚠️ Please install the to ensure uploads and comments are reliably processed by Codecov.

Codecov Report

✅ All modified and coverable lines are covered by tests.
✅ Project coverage is 47.66%. Comparing base (2000379) to head (25bcc6e).
❗ Your organization needs to install the Codecov GitHub app to enable full functionality.

Additional details and impacted files

@@           Coverage Diff           @@
##           master    #6874   +/-   ##
=======================================
  Coverage   47.66%   47.66%           
=======================================
  Files         422      422           
  Lines       52396    52396           
  Branches     7606     7606           
=======================================
  Hits        24973    24973           
  Misses      25632    25632           
  Partials     1791     1791           

Flag	Coverage Δ
go-feature-server	30.58% <ø> (ø)
python-unit	49.00% <100.00%> (ø)

Files with missing lines	Coverage Δ
sdk/python/feast/data_source.py	81.43% <100.00%> (+0.32%)	⬆️

... and 1 file with indirect coverage changes

---

Continue to review full report in Codecov by Harness.

Legend - Click here to learn more
Δ = absolute <relative> (impact), ø = not affected, ? = missing data
Powered by Codecov. Last update 2000379...25bcc6e. Read the comment docs.

🚀 New features to boost your workflow:

• ❄️ Test Analytics: Detect flaky tests, report on failures, and find test suite problems.
• 📦 JS Bundle Analysis: Save yourself from yourself by tracking and limiting bundle sizes in JS merges.