fix(date): import pandas locally in make_ts_exclusive

[anxkhn]

make_ts_exclusive uses pd.Timedelta in its dialect == "tsql" branch, but
pandas is only imported under TYPE_CHECKING in sqlmesh/utils/date.py (it is
banned as a module-level import by the ruff flake8-tidy-imports config). Unlike
the sibling helpers make_inclusive_end and to_utc_timestamp, which do
import pandas as pd at runtime, make_ts_exclusive never imports pandas, so:

from sqlmesh.utils.date import make_ts_exclusive
make_ts_exclusive("2020-01-01 12:00:00", dialect="tsql")
# NameError: name 'pd' is not defined

This is reachable in normal use: rendering an INCREMENTAL_BY_TIME_RANGE model
that has a jinja source/ref event-time filter on the tsql dialect calls
make_ts_exclusive(..., dialect="tsql") in sqlmesh/core/renderer.py. The
NameError is caught there and re-raised as a ConfigError reporting that the
model could not be rendered.

The fix adds a function-local import pandas as pd as the first line of
make_ts_exclusive, mirroring the existing runtime-pandas helpers in the same
module. The non-tsql path is unaffected (it did not reference pd).

Test Plan

• Added test_make_ts_exclusive in tests/utils/test_date.py covering both
  paths: the tsql dialect returns to_utc_timestamp(ts) - 1ns (and no longer
  raises NameError), and a non-tsql dialect (duckdb) returns ts + 1us.
• Verified the test is a true regression test: with the runtime import removed it
  fails with NameError: name 'pd' is not defined; with the fix it passes.
• pytest tests/utils/test_date.py -> 61 passed.
• pytest tests/utils/ -m "not slow and not docker" -> 236 passed, 1 skipped.
• ruff check + ruff format --check + mypy on the changed files -> clean.

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

[anxkhn]

The only failing check here is style-and-cicd-tests (3.9), and the single failure is unrelated to this change:

FAILED tests/core/test_test.py::test_freeze_time_concurrent - sqlmesh.utils.errors.TestError: Failed to create test test_2019_py_model_2 (...) list index out of range

test_freeze_time_concurrent runs the model tests concurrently (concurrent_tasks=8), so this looks like an intermittent concurrency race rather than anything this PR touches. Supporting evidence:

• This PR only adds a function-local import pandas as pd inside make_ts_exclusive (sqlmesh/utils/date.py) plus a unit test in tests/utils/test_date.py; it does not touch the model-testing code path.
• The same job passes on Python 3.10, 3.11, 3.12 and 3.13, on the Windows job, and across all engine-tests-docker (including mssql, the T-SQL engine this fix concerns).
• tests/core/test_test.py::test_freeze_time_concurrent passes consistently for me locally.

Could a maintainer re-run the style-and-cicd-tests (3.9) job? I don't have permission to re-run CI on this repo. Happy to rebase if that's preferred.