feat: Support Spark Streaming trigger modes (Trigger.ProcessingTime) for continuous per-model processing

[mdrakiburrahman]

Summary

Adds processing_time trigger mode that continuously loops models: discover files → process batches → sleep → repeat. Fast models are no longer blocked by slow ones during long dbt run sessions.

Closes #27

Changes

File	What
trigger_config.py	New TriggerConfig frozen dataclass + parse_trigger_config() with interval parser (10s, 1 minute, etc.)
constants.py	DEFAULT_PROCESSING_TIME_TIMEOUT_SECONDS (30 days)
impl.py	wait_for_next_cycle() (interruptible sleep via threading.Event), parse_trigger(), reset_cycle_count(), get_processing_time_timeout(), SIGTERM/SIGINT signal handlers
incremental.sql	Trigger config reading, dynamic loop cap (999M for processing_time), inter-cycle sleep + re-discover, auto 30-day timeout, cache clearing between cycles
README.md	New "Streaming / Continuous Processing" section
test_trigger_config.py	27 tests — parsing, validation, interval formats, error cases
test_trigger_lifecycle.py	11 tests — cycle counting, max_cycles, shutdown signals, interruptible sleep

Usage

-- Continuous processing: loop forever, sleeping 30s between cycles
{{ config(
    materialized='incremental',
    incremental_strategy='append',
    trigger={'type': 'processing_time', 'interval': '30 seconds'},
    source_roots=['/shares/...'],
    ...
) }}

-- With max_cycles for controlled shutdown
{{ config(
    trigger={'type': 'processing_time', 'interval': '10 seconds', 'max_cycles': 100},
    ...
) }}

Design Decisions

• Batch loop stays in Jinja — processing_time uses a very large loop cap (999,999,999). When files are exhausted, Jinja calls adapter.wait_for_next_cycle() which sleeps and returns True/False.
• Only incremental materialization supports processing_time (table does full DELETE each time, incompatible with continuous processing).
• threading.Event for graceful shutdown — thread-safe, interruptible sleep via event.wait(timeout=interval).
• Auto 30-day timeout for processing_time models (user can override via job_timeout_seconds).

Testing

• ✅ 335 unit tests pass (including 38 new trigger tests)
• ✅ 3 integration tests pass (full refresh, incremental, filtered edition)
• ✅ Lint clean (ruff check + format)