fix(cdp): pass schema as runtime option and add typed join handling

[adi-IL]

Fixes #88, Fixes #89

What was wrong

1. In pipelines/cdp/cdp_pipeline/customer_data_platform.py, the BigQuery output schema was read directly at module import time with a hardcoded relative path (with open("./schema/unified_table.json", encoding="utf-8")). When the package was installed on Dataflow workers or executed outside the pipelines/cdp directory, the relative path failed with FileNotFoundError.
2. In left_join, coupon_redempt_value lacked proper type annotations and type checks, triggering linter warnings regarding dictionary subscriptability on Optional types.

What changed

• Removed top-level schema file loading from module scope.
• Added load_output_schema(schema_path) helper function that supports custom schema paths, package-relative fallback paths, and embedded default schema definitions.
• Added --output_schema_path argument to MyPipelineOptions in pipelines/cdp/cdp_pipeline/options.py.
• Updated create_and_run_pipeline to load and pass the schema dynamically at pipeline launch time.
• Added complete type annotations and defensive dictionary checks to left_join for clean linter validation.
• Added unit tests in pipelines/cdp/tests/test_customer_data_platform.py covering matching joins, non-matching joins, schema loading, and Beam pipeline transformation execution with TestPipeline.

Verification

• yapf -i -r --style yapf . -> clean formatting
• pylint --rcfile ../pylintrc cdp_pipeline tests -> 10.00/10 rating
• pytest tests/ -> 4 passed
• python setup.py sdist -> package builds successfully

[iht]

LGTM! Verified that both #88 (schema loading at launch time with package fallback and CLI option) and #89 (type annotations, defensive dictionary handling in left_join, and unit tests) are fully resolved. All CI checks passed successfully.