fix: normalize internal state timestamps

[nicosuave]

Summary

• Normalize dlt internal metadata timestamp columns before Iceberg schema inference/evolution.
• Keep _dlt_pipeline_state.created_at as Iceberg timestamp when incoming Arrow batches are timestamp[us, tz=UTC].
• Align _dlt_loads.inserted_at with the existing naive UTC TimestampType convention used by _dlt_version.

Addresses #9.

Tests

• uv run pytest tests/test_with_state_sync.py::TestWithStateSyncMethods::test_state_table_created_at_timestamptz_batch_matches_existing_timestamp_schema -q
• uv run pytest tests/test_with_state_sync.py tests/test_load_metadata_resilience.py tests/test_schema_evolution.py -q
• uv run pytest tests/test_class_based_atomic.py tests/test_e2e_sqlite_catalog.py tests/test_destination_e2e.py -q
• uv run pytest tests/test_schema_converter.py tests/test_schema_casting.py tests/test_state_sync_e2e.py -q
• uv run pytest -q -m "not integration" (148 passed, 15 deselected)

[chatgpt-codex-connector]

💡 Codex Review

Here are some automated review suggestions for this pull request.

Reviewed commit: 5e303bc7c1

ℹ️ About Codex in GitHub

Your team has set up Codex to review pull requests in this repo. Reviews are triggered when you

• Open a pull request for review
• Mark a draft as ready
• Comment "@codex review".

If Codex has suggestions, it will comment; otherwise it will react with 👍.

Codex can also answer questions or update the PR. Try commenting "@codex address that feedback".

[chatgpt-codex-connector]

Preserve existing timestamptz metadata tables

When an internal metadata table was already created by the previous code with Iceberg TimestamptzType (for example _dlt_pipeline_state.created_at from an earlier timestamp[us, tz=UTC] batch), this unconditional normalization turns the next incoming batch into TimestampType before schema evolution runs. The existing-table path then compares TimestamptzType to TimestampType and evolve_schema_if_needed rejects that as an unsafe type change, so affected pipelines fail on the first run after upgrading instead of simply casting to the existing schema. Consider only normalizing when the existing column is already TimestampType, or treating the internal timestamp/timestamptz pair as compatible during evolution.

Useful? React with 👍 / 👎.