fix(analyzer): report channel-aware MDR outcomes

[sent-dm]

What changed

• Normalized every MDR row into channel, direction, observed statuses, terminal outcome, diagnostics, and optional ERR_* codes.
• Grouped results by channel and direction while retaining explicit unknown buckets.
• Separated progression, terminal failure, deferred, inbound, malformed, and unknown outcomes with globally reconciling totals.
• Calculated transitions only from explicit histories; latest-status-only records no longer backfill unobserved stages.
• Stopped SMS delivery at DELIVERED and reported WhatsApp/RCS READ as engagement.
• Added stable --format json output with null denominators, text N/A, and exit 2 when no usable cohort exists.
• Replaced MDR fixtures and updated instructions, references, routing evals, and utility tests.

Why

The previous analyzer treated every latest status as proof of all earlier transitions and applied READ as a universal stage. That distorted SMS results and could label all-unknown inputs healthy.

Validation

• python3 scripts/generate_adapters.py --check
• python3 scripts/validate.py
• python3 scripts/test_fixtures.py
• python3 scripts/test_contracts.py
• python3 scripts/test_validation_gates.py
• git diff --check

Stacked on #5.