Pin tolerance for additive sample fields and explicit-null samples

[mbreissi]

What

Regression pins (tests only, no pipeline changes) for the amended SOUTHBOUND §2 conventions — "Explicit null values" and "Additive sample fields" (core/docs/SOUTHBOUND.md):

• A samples[] entry MAY carry additive protocol-specific fields beside the canonical five (e.g. valueType/valueEncoding); consumers MUST ignore fields they do not understand.
• A deliberate null publishes as value: null with quality: "GOOD" (a failed read is value: null + quality: "BAD"), so the two stay distinguishable by quality alone.

The telemetry-processor is payload-agnostic by design; these tests pin that the §2 shapes flow through ingestion and every representative pipeline stage without error, without dropping the sample, and — for pass-through/projection paths — without stripping the additive fields.

Tests added

Test	Pins
dispatch::tests::forwards_additive_sample_fields_and_explicit_null_value_untouched	ingestion fan-out forwards the body verbatim (null + extras intact, sample not dropped)
proc::filter::tests::quality_filter_keeps_explicit_null_value_with_good_quality_and_additive_fields	quality filter keys on quality only: null+GOOD kept, null+BAD droppable
proc::project::tests::keep_preserves_additive_sample_fields_and_explicit_null_value	project.keep copies samples byte-identical, extras included
proc::tests::southbound_s2_null_and_additive_sample_fields_flow_through_a_pass_through_pipeline	full filter → sample → project pipeline carries the sample end to end verbatim
proc::aggregate::tests::explicit_null_samples_fold_per_documented_null_semantics	aggregation does not crash on nulls and follows its documented semantics

Null-handling findings

The aggregate stage has documented null semantics (docs/reference/data-types.md "Aggregate agg value types", docs/reference/messaging-interface.md "Aggregate output"), and the implementation matches them — no gap found:

• count counts all folded sample values, nulls included;
• first/last carry the raw value, null included;
• numeric reducers (avg/sum/min/max) fold numeric values only and emit null when the window had no numeric sample (an all-null window emits samples[0].value: null, quality: "GOOD" — consistent with §2''s explicit-null convention).

No pipeline or dispatch code changed. Full suite: 99 passed, 0 failed (baseline was green).

🤖 Generated with Claude Code

https://claude.ai/code/session_014iQDskU2E1HLStYHoZrSTn