fix(dlt): use timestamp macros for incremental model time filter

[anxkhn]

Description

The DLT integration generates INCREMENTAL_BY_TIME_RANGE models whose time column is cast to a timestamp:

TO_TIMESTAMP(CAST(c._dlt_load_id AS DOUBLE)) as _dlt_load_time

but it filtered that column with the categorical date macros:

WHERE
  TO_TIMESTAMP(CAST(c._dlt_load_id AS DOUBLE)) BETWEEN @start_ds AND @end_ds

@start_ds / @end_ds render bare dates (midnight on both ends), while make_inclusive (sqlmesh/utils/date.py) defines the inclusive range as 00:00:00 .. 23:59:59.999999. The timestamp form of that range is what @start_ts / @end_ts emit. On a single-day run (@start_ds == @end_ds) both bounds collapse to midnight, so any row with a non-midnight timestamp (e.g. 21:30:15) is excluded and the result is empty.

This matches the diagnosis and suggested fix in #5689: filter a timestamp column with @start_ts AND @end_ts.

Fixes #5689

Test Plan

• Added tests/integrations/test_dlt.py: a focused unit test that calls generate_incremental_model(...) and asserts the generated WHERE uses BETWEEN @start_ts AND @end_ts (and no longer @start_ds / @end_ds). No warehouse or dlt package needed. Confirmed failing-first: reverting the source line back to @start_ds / @end_ds makes it fail; the fix makes it pass.
• Updated the four hard-coded expected-SQL strings in tests/cli/test_cli.py (the DLT pipeline template tests). test_dlt_filesystem_pipeline, test_dlt_pipeline, test_dlt_pipeline_errors -> 3 passed (dlt 1.28.1, filesystem destination, DuckDB default, no warehouse).
• make style (ruff format/check + mypy) clean on changed files.

Checklist

• I have run make style and fixed any issues
• I have added tests for my changes (if applicable)
• All existing tests pass (make fast-test)
• My commits are signed off (git commit -s) per the DCO