feat: add --threads option to parallelize report data fetching

Add --threads CLI option (default 1) to `edr report` and `edr send-report`.
When set to >1, independent dbt run-operations are executed concurrently
using ThreadPoolExecutor with SubprocessDbtRunner.

dbt's Python API (APIDbtRunner) is not thread-safe due to global mutable
state (GLOBAL_FLAGS, adapter FACTORY, etc.), so parallel execution uses
SubprocessDbtRunner which spawns independent dbt processes per call.

The fetching is split into phases:
- Phase 1: 14 independent operations run in parallel
- Phase 2: exposures + test_results (depend on Phase 1)
- Phase 3: lineage (depends on Phase 2)
- Phase 4: pure computation (no dbt calls)

With --threads=14, edr report time is expected to drop from ~3m40s to
~30-40s on adapters with high query latency (e.g. Athena).

Author: dtaniwaki