fix(core): align SQL end_date filter with whole-day semantics

[Sanjays2402]

Description

TaskQueryBuilder._build_date_filter_conditions compared the datetime columns against the bare end_date, which SQLite coerces to midnight, so a same-day value carrying a time was excluded. The default TaskRepository.get_filtered() truncates to a date and includes it, so both repository implementations returned different results for the same query. end_date is now turned into an exclusive next-day bound, making both paths whole-day inclusive.

Related Issue

Closes #1164

Type of Change

• Bug fix (non-breaking change which fixes an issue)

Changes Made

• _build_date_filter_conditions: end_date becomes an exclusive end_date + 1 day bound (field >= start / field < end + 1d) instead of between/<= end_date.
• TaskRepository._matches_date_filter: docstring now names the SQL helper it mirrors and states the whole-day-inclusive end_date semantics.
• New regression test with a time-bearing boundary value asserting both implementations agree; updated the query-builder SQL-shape assertion that checked for between.

Testing

Test Environment

• OS: macOS
• Python version: 3.14
• UV version: N/A (plain venv + pytest)

Tests Performed

• Unit tests pass (packages/taskdog-core/tests, 1175 passed)
• Added new tests for new features
• Manual testing completed

test_date_filter_end_date_includes_same_day_time fails on main (assert 0 == 1 — the task is dropped by the SQL path) and passes with the fix; verified by stashing only the source change.

Code Quality Checklist

• Linter passes (Ruff check on changed files)
• Type checker passes (make typecheck)
• Code is formatted (make format)
• No new warnings introduced
• Code follows project conventions

Documentation

• Added/updated docstrings

Package Affected

• taskdog-core

Breaking Changes

Behavioural change to the SQL filter only, bringing it in line with the documented inclusive end_date and with the default repository implementation.

Checklist

• My code follows the coding standards of this project
• I have performed a self-review of my own code
• I have commented my code in hard-to-understand areas
• My changes generate no new warnings
• I have added tests that prove my fix is effective
• New and existing unit tests pass locally with my changes

[Kohei-Wada]

Thanks for the fix — the diagnosis is right and the direction is correct. end_date is documented as inclusive, so making the SQL path whole-day inclusive (rather than making the default repository exclusive) is the right way to reconcile the two implementations.

I verified locally: the new test_date_filter_end_date_includes_same_day_time fails on main with assert 0 == 1 when only the source change is reverted, and the full taskdog-core suite passes with the fix (1175 passed).

Two minor nits in _build_date_filter_conditions:

1. end_bound is recomputed on every iteration of the for field in date_fields loop. It doesn't depend on field, so please hoist it above the loop.

2. (field >= start_date) & (field < end_bound) — the surrounding code uses the SQLAlchemy function style (or_(...)), so and_(field >= start_date, field < end_bound) would be more consistent, and it may also let you drop the # type: ignore[operator].

Neither affects behaviour. Happy to merge once these are addressed.

[Kohei-Wada]

Thanks! Nice catch on the two repository implementations disagreeing — the regression test asserting both paths agree is exactly the right shape. Merging.