fix(otlp): complete post-#262 DDSketch encoding-value + proto wire-type absorption

[zzylol]

Summary

Sweep blocker #1: completes the post-PR-#262 absorption that left
EXPORTER_SDK_AGG=dd-delta rejected at the OTLP exporter and
EXPORTER_SDK_AGG=dd-full rejected at the agent / backend decoder.
Without this, none of run_e2e_sweep.sh's DDSketch matrix cells run,
blocking paper claims ① ④ ⑤.

The PR delivers Fix 1 (encoding-string mapping) cleanly and pins the
agent-decoder wire-format invariants with regression tests. Fixes 2

• 3 share a structural root cause that requires a sketchlib-go
  upstream migration outside this PR's touch list — see
  "User-attention" below.

Per-failure root cause

Failure 1 — unknown ddsketch encoding: ddsketch_proto_delta

Root cause: DDSketchEncodingValue in
exporters/otlp/otlpmetric/{otlpmetricgrpc,otlpmetrichttp}/internal/ transform/metricdata.go only matched
metricdata.DDSketchEncodingProto. PR #262 added
metricdata.DDSketchEncodingProtoDelta (+ the mpb.DDSketchEncoding_ DDSKETCH_ENCODING_PROTO_DELTA proto enum) but the encoding-switch
arm was not extended, so any data point carrying the new enum hit
the default-case errUnknownDDSketchEncoding.

Fix: added the missing arm in both exporters and the shared
template (internal/shared/otlp/otlpmetric/transform/ metricdata.go.tmpl). Sibling sketches (HLL / CountSketch /
CountMinSketch) were already absorbed in #262; only DDSketch was
missed. KLL has no delta path, so its single-arm switch is correct.

Failure 2 — DdSketchState.alpha: invalid wire type: LengthDelimited (expected SixtyFourBit) + DDSketch SketchEnvelope did not carry a DDSketchState variant

Root cause (single, structural): the SDK's DDSketch aggregator
(opentelemetry-go-patch/sdk/metric/internal/aggregate/ddsketch.go)
serializes via proto.Marshal(sk.ToProto()) against
github.com/DataDog/sketches-go/ddsketch/pb/sketchpb.DDSketch, but
every consumer (asap-precompute-go's DDSketchWrapper.ApplyDelta,
asap-precompute-rs's DDSketchWrapper::decode_envelope, the agent
processor's decodeDDSketchEnvelope) expects sketchlib-go's
SketchEnvelope{DDSketchState}. The two schemas collide on field 1:
DataDog's Mapping is a sub-message (LengthDelimited wire type)
while sketchlib-go's alpha is double (SixtyFourBit). When the
consumer's proto.Unmarshal hits the field-1 wire-type clash it
emits the reported error; when it doesn't, env.GetDdsketch()
returns nil (no oneof variant set in DataDog bytes) and surfaces
"did not carry a DDSketchState variant".

KLL / HLL / CountSketch / CountMinSketch already migrated to
sketchlib-go's portable format (see PR #262 §"sketchlib-go API
rename refactor (PR #53) absorption"); DDSketch remained on DataDog.
The right structural fix is to migrate DDSketch's aggregator to
sketchlib-go's *ddsketch.DDSketch and serialize via
SerializePortable + proto.Marshal. This sits outside the
sweep-blocker-1 touch list — see User-attention below.

Test additions (3 new test files, 12 new test functions)

1. opentelemetry-go-patch/exporters/otlp/otlpmetric/otlpmetricgrpc/ internal/transform/encoding_value_test.go (mirrored to the
   otlpmetrichttp sibling): 7 test functions
   (TestDDSketchEncodingValue, TestDDSketchEncodingValue_Unknown,
   TestKLLSketchEncodingValue_Proto,
   TestCountSketchEncodingValue,
   TestCountMinSketchEncodingValue, TestHLLSketchEncodingValue,
   TestDDSketchDataPointsCarriesProtoDeltaEncoding) covering both
   the new ProtoDelta arm and the sibling-sketch surface to catch
   future re-introductions of the same drift.
2. opentelemetry-collector-contrib-patch/processor/ddsketchprocessor/ envelope_decode_test.go: 3 test functions
   (TestDecodeDDSketchEnvelope_RoundTripFull,
   TestDecodeDDSketchEnvelope_BareStateRejected,
   TestDecodeDDSketchEnvelope_EmptyEnvelopeRejected) pinning the
   agent decoder's wire-format contract.

Sibling sketches — open question

None left. The post-#262 absorption gap was DDSketch-only.
HLL / CountSketch / CountMinSketch all have working Delta
encoding-value arms verified by the new sibling tests. KLL has no
delta path, which is correct.

Doc-tightening

No stale post-rename comments referencing pre-rename API names
(Add / InsertValue / GetValueAtQuantile etc.) found in any
file touched by this PR. The remaining series.sketch.Add(...)
call in internal/aggregate/ddsketch.go is DataDog's API, not
sketchlib-go's pre-rename API; it'll be replaced when the
structural migration lands.

User attention required

This PR delivers Fix 1 fully and pins decoder invariants. Fixes 2
and 3 require user attention because they share a structural root
cause that can't be resolved within sweep-blocker-1's touch list:

• The alpha wire-type mismatch is structural — the proto schema
  (sketchlib-go's DDSketchState) and the SDK encoder (DataDog's
  sketchpb.DDSketch) cannot be aligned without a semantic change.
• The fix needs a sketchlib-go ↔ SDK migration: rewrite
  opentelemetry-go-patch/sdk/metric/internal/aggregate/ddsketch.go
  to use github.com/ProjectASAP/sketchlib-go/sketches/DDSketch (as
  KLL/HLL/CountSketch/CountMinSketch already do) and emit via
  SerializePortable + proto.Marshal for full state and
  ddsketch.ComputeDelta for the delta encoding. That file is on
  the sweep-blocker-1 do-not-touch boundary (the touch list scopes
  to encoding-value map + OTLP encoding switch + processor decoder
  • Rust wrapper).

Test plan

• go test ./internal/transform/... green for otlpmetricgrpc
  + otlpmetrichttp exporters.
• go test ./... green for the patched
  opentelemetry-collector-contrib/processor/ddsketchprocessor.
• go test ./... green for opentelemetry-go/sdk/metric/....
• cargo test --release green for asap-precompute-rs (no
  Rust changes; parity tests pass).
• (Out of this PR's scope) Run run_e2e_sweep.sh after the
  structural follow-up lands to confirm the matrix's DDSketch
  cells produce real PromQL answers.

🤖 Generated with Claude Code