otap: implement Schema/Dictionary/Record codec from data_model.md

[zzylol]

Summary

An audit against docs/data_model.md found that the OTAP Strategy-B wire codec (encode_batch/decode_batch) never actually implemented the doc's Schema/Dictionary/Record economics: every row inlined metric, labels, sketch_type, agg_id, schema_version, and encoding, repeated on every RECORD row, every window, forever — the exact anti-pattern the doc's rationale exists to avoid. The design only ever existed as a field-categorization scheme, not an implemented protocol behavior.

This PR implements it for real.

What changed

• New otap::dictionary module: SeriesDictionary (sender) assigns stable series_ids and emits SCHEMA/DICTIONARY/LABELS rows only the first time an agg_id/series is seen; RECORD always carries just series_id + window bounds + envelope/value. SeriesDictionaryDecoder (receiver) retains that state across a continuous stream and reconstructs full SketchEnvelopes by joining RECORD rows back against it — an unresolvable series_id/agg_id is a hard decode error, not a silent partial result, per the doc's own statefulness caveat.
• sketch_size is resolved from PrecomputeConfig::sketch_params via a new Precompute::active_config() trait method.
• hash_seed/hash_function resolve asap_sketchlib's HashSpec down to the one canonical seed position (seed_list[canonical_seed_index]) rather than carrying the full 20-entry seed table — checked against asap_sketchlib's own self-describing wire-format doc, since one SCHEMA row (one agg_id, one sketch_type) only ever needs one seed position. SeriesDictionaryDecoder::schema_for() exposes the resolved value on the receiver side without fabricating a lossy reconstructed HashSpec.
• StubPlugin and AsapSketchesPlugin now use this codec for tick/drain (encode) and inbound-envelope (decode), each with persistent dictionary state across calls.
• encode_batch/decode_batch/records::{flatten,lift} are unchanged — they solve a different problem (disguising a payload as OTAP-Metrics-shaped so it can transit a generic OTAP pipeline hop) that docs/data_model.md was never about (its own first line scopes it to the asap_sketches-to-asap_sketches hop).
• New examples/sketch_pipeline_demo.rs: a runnable three-stage demo — sketch creation processor → receive processor (merges via observe_envelope, queries via estimate mode) → Prometheus text exposition. Run with cargo run --example sketch_pipeline_demo --features otap. The printed output makes the dictionary economics directly visible: window 0 sends all four batches, every later window sends RECORD only.

Testing

• cargo test --features otap: 149/149 passing (14 new tests covering first-window-vs-repeat-window batch shapes, distinct series_id assignment, full encode→decode round trips, hash-seed resolution, and decode error handling for unknown series_id/agg_id).
• cargo fmt --check and cargo clippy --features otap --tests --examples: clean.
• cargo run --example sketch_pipeline_demo --features otap: verified end to end — window 0 emits schema=1 dictionary=1 labels=1 record=1, windows 1-3 emit schema=0 dictionary=0 labels=0 record=1, and the receiver's queried p99 gauge tracks the injected latency drift correctly.

🤖 Generated with Claude Code