fix(processors): unbreak baselines - DataDog->sketchlib-go fixtures + post-#211 sink assertions

[zzylol]

Summary

Restores green baselines across all 5 OTel sketch processors (ddsketch, kll, hll, countsketch, countminsketch) so refactor work can proceed without inheriting pre-existing debt.

Why

Three classes of pre-existing test failure, one per cluster of processors:

1. DataDog/sketches-go fixture leftover (ddsketch). Production switched to sketchlib-go in fix(ddsketch): make delta_transmission actually work end-to-end #210; tests still imported DataDog/sketches-go to build sketch fixtures. Done in a prior commit (already on this branch): import swap to github.com/ProjectASAP/sketchlib-go/sketches/DDSketch and proto/sketch_envelope; setSketchPayload uses ddsketch.NewDDSketch + Update + SerializeStateProtoBytes; decodeSketch tries envelope first then bare-state fallback.

2. Post-fix(processors): forward inputs through windowed sketch processors #211 window-mode pass-through. After PR fix(processors): forward inputs through windowed sketch processors #211, window-mode ConsumeMetrics forwards raw inputs through nextConsumer so chained windowed sketch processors (processors: [kll, hll, batch]) can see the original payload. Tests written before fix(processors): forward inputs through windowed sketch processors #211 expected sink.AllMetrics() to be empty after ConsumeMetrics; they now observe both the input pass-through AND the synthesized output emitted by flushWindow. Updated assertions to count N inputs + 1 flush = N+1 entries and scan-by-name where ordering is non-deterministic (concurrent test).

3. selfmonitor import + go.mod replace missing (kll, countsketch, countminsketch). These processors import go.opentelemetry.io/collector/processor/selfmonitor, a package that lives in the local fork at opentelemetry-collector/processor/selfmonitor but is not in any upstream tag. The corresponding go.mod was missing the replace go.opentelemetry.io/collector/processor => ../../../opentelemetry-collector/processor directive that ddsketchprocessor already had. Added the replace; go mod tidy resolves cleanly afterward.

A bonus issue surfaced for kll/countminsketch: production emits sketch payloads via proto envelopes (SerializePortable + proto.Marshal for KLL, SerializeProtoBytesFO for CMS), but two tests called Deserialize<sketch>FromBytes (gob-encoded) — which mismatches the wire format and panics with gob: duplicate type received. Switched the tests to Deserialize<sketch>FromProtoBytes to match the emit path.

Scope

Per the user-supplied plan, this PR does NOT touch:

• opentelemetry-collector-contrib/ (unmodified upstream mirror)
• telegraf-patch/plugins/aggregators/ddsketch/ (separate DataDog cleanup; out of scope)
• cmd/telemetrygen/ (separate DataDog cleanup; out of scope)
• asap-precompute-go/ (already merged in feat(asap-precompute-go): bootstrap module + types + window + snapshot cache + OTel adapter (Phase 2 steps 2.1-2.4) #219)

Submodule pointer drift on opentelemetry-collector and opentelemetry-go was left unstaged (pre-existing noise).

Per-processor results

Processor	go test	go build	gofmt (modified files)
ddsketchprocessor	pass (no fail)	pass	clean
kllprocessor	pass (no fail)	pass	clean
hllprocessor	pass (no fail)	pass	clean
countsketchprocessor	pass (no fail)	pass	clean
countminsketchprocessor	pass (no fail)	pass	clean

(gofmt issues remain in files I did not touch — pre-existing debt, out of scope.)

Full collector build via ./build_sketchcollector.sh succeeds end-to-end: builder produces cmd/sketchcollector/sketchcollector binary.

Files modified per processor

• ddsketchprocessor: processor_test.go (5 window-mode tests + concurrent test → expect input pass-through + flush; index synthesized output by name where ordering is non-deterministic), go.mod / go.sum (DataDog dep removed in prior commit on this branch).
• kllprocessor: processor_test.go (5 window-mode tests fixed; gob→proto deserializer in TestBatchModeTransmitSketch), go.mod (added replace go.opentelemetry.io/collector/processor).
• hllprocessor: processor_test.go (2 window-mode tests fixed), go.mod / go.sum (no replace needed; go mod tidy rebalanced direct vs. indirect).
• countsketchprocessor: go.mod / go.sum only — added replace go.opentelemetry.io/collector/processor. Tests already accounted for pass-through.
• countminsketchprocessor: processor_test.go (gob→proto deserializer in TestProcessor_TumblingWindow_Correctness), go.mod (added the replace).

Test plan

• cd <each-processor> then go mod tidy && go build ./... && go test ./... -count=1 — all 5 pass.
• ./build_sketchcollector.sh — succeeds.
• Per code-review: assertion changes truthfully reflect the post-fix(processors): forward inputs through windowed sketch processors #211 pass-through contract; no test was weakened (all original semantic checks preserved, just relocated to scan the synthesized entry instead of out[0]).