fix(execution): make ERA collapse ordering deterministic

[egillax]

Summary

Fixes nondeterministic final ERA collapse in the Ibis execution engine.

The collapse window previously ordered only by start_date. When multiple intervals had the same start
date, the backend could process the shorter interval before the longer one. That could leave a later
interval, fully contained by the longer interval, as an extra cohort row.

This change makes the collapse ordering deterministic by processing longer tied-start intervals first.

Minimal Reprex

Input intervals:

2020-01-01 -> 2020-01-02
2020-01-01 -> 2020-02-01
2020-01-10 -> 2020-01-15

Before:

2020-01-01 -> 2020-02-01
2020-01-10 -> 2020-01-15

After:

2020-01-01 -> 2020-02-01

The second Before row is contained in the first row and should not survive ERA collapse.

## Validation

uv run --extra dev pytest tests/execution/
test_end_strategy_censoring.py::test_collapse_settings_era_merges_contained_intervals_after_tied_start_d
ates -q
uv run --extra dev pytest tests/execution -q

Results:

1 passed
271 passed

Private real-data validation against the first 100 PhenotypeLibrary cohorts on real data reduced
CirceR row count mismatches from 19/100 to 0/100.

[codecov]

Codecov Report

✅ All modified and coverable lines are covered by tests.
✅ Project coverage is 85.79%. Comparing base (76466c5) to head (b4d9419).

Additional details and impacted files

@@           Coverage Diff            @@
##           develop      #36   +/-   ##
========================================
  Coverage    85.79%   85.79%           
========================================
  Files          169      169           
  Lines        12514    12514           
========================================
  Hits         10737    10737           
  Misses        1777     1777           

☔ View full report in Codecov by Sentry.
📢 Have feedback on the report? Share it here.

🚀 New features to boost your workflow:

• ❄️ Test Analytics: Detect flaky tests, report on failures, and find test suite problems.

[azimov]

No issue with the fix - I've tested and confirmed this.

I will create an issue for better testing to have higher visibility for more real cohorts between ibis and java implementations.