New span source: ingest nested OTLP resourceSpans/scopeSpans exports

[bamdadd]

Ingest today (src/taintline/ingest.py, _parse_otlp) expects a flat JSON list of span dicts with camelCase keys. Real OTLP JSON exporters (the OpenTelemetry Collector file exporter, otlp/json) emit the nested envelope resourceSpans[].scopeSpans[].spans[], where each spans entry carries spanId/parentSpanId as hex strings and attributes as an array of {key, value:{stringValue|intValue|boolValue|...}} objects rather than a flat map. taintline cannot read those exports as-is. Ingest stays pure and deterministic: no network, clock, or RNG.

Acceptance criteria

• load_trace auto-detects the nested OTLP envelope (by structure) and flattens resourceSpans[].scopeSpans[].spans[] into the existing normalized Span/Trace types with no changes to model.py.
• OTLP attributes key/value arrays (including stringValue, intValue, boolValue, arrayValue) are decoded into the flat attributes dict so the existing gen_ai.* key lookups keep working; typed helpers, mypy strict, no loose Any-soup.
• Existing flat-list and Strands inputs still parse unchanged (open-closed: add a branch, do not rewrite the current paths).
• A portability fixture under tests/fixtures/portability/ holds a small real-shaped nested OTLP export and a test asserts the same Verdict the flat form would produce. Domain-neutral tool names.
• Determinism preserved: same file -> byte-identical Trace/Verdict across runs.

[IronLad123]

I'll take this on! I will implement auto-detection and normalization for nested OTLP resourceSpans[].scopeSpans[].spans[] envelopes in src/taintline/ingest.py, decode typed OTLP attributes arrays into flat attribute maps, preserve existing flat-list and Strands parsing paths, add portable test fixtures, and verify determinism.