feat: support starting_timestamp to start from an offset

[mdrakiburrahman]

Why this change is needed

Closes #11

When a model has no checkpoint yet, dbt-scope processes all source files from the beginning of time. For large datasets with years of historical files, this can be very expensive. Users need a way to skip historical files and start processing from a given point in time — analogous to Delta Streaming's startingTimestamp.

How

A new starting_timestamp config parameter (ISO-8601 UTC string, e.g. "2026-04-07T10:00:00+00:00") is added to the adapter's file discovery pipeline.

Behavior:

Checkpoint exists?	starting_timestamp set?	Behavior
Yes	Any	No-op — checkpoint watermark takes precedence
No	Not set	Process all files (existing behavior, backward compatible)
No	Valid, before latest file	Skip files ≤ timestamp, process the rest
No	Valid, after all files	Throw DbtRuntimeError
No	Invalid/unparseable	Throw DbtRuntimeError
No	Naive (no timezone)	Throw DbtRuntimeError

Key implementation details:

• _parse_starting_timestamp() validates the timestamp string and ensures it has timezone info, converting to UTC
• When no checkpoint exists and starting_timestamp is provided, a synthetic Watermark is created from it and used for filtering — no changes to file_tracker.py or checkpoint.py were needed
• The "timestamp after all files" check only triggers an extra LIST call in the error case (no performance impact on the happy path)
• Both table.sql and incremental.sql macros read the config and pass it through to adapter.discover_files()

Files changed:

• dbt/adapters/scope/impl.py — core logic: _parse_starting_timestamp(), extended discover_files() and has_unprocessed_files()
• dbt/include/scope/macros/materializations/table.sql — reads and passes starting_timestamp
• dbt/include/scope/macros/materializations/incremental.sql — reads and passes starting_timestamp
• tests/integration/dbt_project/models/append_no_delete.sql — uses 1900-01-01 (process all)
• tests/integration/dbt_project/models/filtered_edition.sql — uses 2026-01-01 (skip pre-2026)

Test

• 12 new unit tests in tests/unit/test_starting_timestamp.py covering:
  • Timestamp parsing: valid UTC, valid with offset, invalid string, empty string, naive (no tz)
  • Filtering: used when no watermark, ignored when watermark exists, correct file filtering
  • Validation: invalid raises, after-all-files raises, empty source returns empty, None processes all
• All 210 unit tests pass (uv run pytest tests/unit/ -q)
• Ruff lint and format checks pass on all changed files