refactor(module): relocate patterns + replay into module/pkg/ (PR-I.2a)

[trilamsr]

Summary

Mechanical reshuffle of the pattern library and replay runner from
internal/synthesis/{patterns,replay}/ into module/pkg/{patterns,replay}/
so the moat surface lives inside the same Go submodule that hosts the OTel
components PR-I.2b will introduce (rankjoinprocessor,
patterndetectorprocessor).

This is PR-I.2a per the RFC-0013 §migration L249 sub-slice; net-new
processor packages, builder-config.yaml wiring, and the
module/v0.2.0 tag bump land in PR-I.2b later.

Why

internal/synthesis/patterns/ was the only piece of the differentiator
("moat") still living in the root module's internal/ tree after the
PR-K wave deletions. Once PR-I.2b introduces the processor packages
they will import the pattern lib from module/; a cross-module import
chain that today is module/receiver/* → root/internal/synthesis/* is
the wrong direction for the post-pivot architecture (processors target
the OCB-distribution baseline at collector v0.110.0 which module/
pins; root tracks v1.59.0).

What moved

internal/synthesis/patterns/  →  module/pkg/patterns/
internal/synthesis/replay/    →  module/pkg/replay/

24 files renamed via git mv (rename similarity 93-100%), zero content
edits except:

• 5 Go files: import-path bump
  github.com/tracecoreai/tracecore/internal/synthesis/{patterns,replay}
  → …/module/pkg/{patterns,replay}.
• 4 doc-comment path mentions inside the moved files (point at the new
  home).
• module/pkg/replay/pod_evicted/_real_world/README.md: 3 path
  references updated.

Module-graph plumbing

• module/go.mod gains direct requires on
  github.com/santhosh-tekuri/jsonschema/v6 v6.0.2 and
  github.com/stretchr/testify v1.9.0 (the moved test files' deps);
  module/go.sum picks up dlclark/regexp2 v1.11.0 (jsonschema's
  transitive).
• Root go.mod flips santhosh-tekuri/jsonschema/v6 to // indirect
  (root-module Go code no longer references it directly; the only
  consumer was verdict_schema_test.go, which moved).
• MVS-trap honored: module/go.mod's collector pins stay at
  v0.110.0 after go mod tidy. The OCB-distribution baseline pin
  from PR-I.1b is preserved — scraperhelper is still part of
  collector/receiver at v0.110.0 and would break the
  hostmetricsreceiver@v0.110.0 build if it floated forward to
  v1.59.0.

CI / workflow path updates

• .github/workflows/bench.yml (M19 benchmark) →
  ./module/pkg/patterns/.
• .github/workflows/chaos.yml paths-filter trigger:
  internal/synthesis/** → module/pkg/patterns/** +
  module/pkg/replay/** (deliberately tighter than module/pkg/**
  so unrelated module additions, e.g. nccl_fr_parser, do not fire the
  chaos gate).
• .github/workflows/chaos.yml M19 race-test invocation:
  go test ./internal/synthesis/... →
  cd module && go test ./pkg/patterns/... ./pkg/replay/....
  The naive rewrite ./module/pkg/... does not resolve from root
  — module/ is its own Go module; the root ./... glob does not
  cross into it under either GOWORK=on or GOWORK=off. Verified the
  fixed invocation passes with -race -count=1.

Doc references updated

• docs/rfcs/0013-distro-first-pivot.md PR-I.2 paragraph amended to
  describe the PR-I.2a / PR-I.2b sub-slice.
• docs/followups/{M19,M4b,otlphttp}.md,
  docs/research/m16-kueue-production-followups.md: forward-looking
  references repointed at module/pkg/.
• Historical RFC + CHANGELOG entries that stamp prior receiver-era
  paths are intentionally left intact.

Commits

1. refactor(module): relocate patterns + replay into module/pkg/ (PR-I.2a)
   — the 24-file git mv + import-path bump + go.mod/go.sum tidy +
   forward-looking doc + bench.yml + initial chaos.yml update.
2. ci(chaos): scope path triggers + test invocation to patterns + replay
   — adversarial-review follow-up tightening chaos.yml paths-filter
   and fixing the broken root-module test invocation.

Out of scope (do not look for these)

• No rankjoinprocessor or patterndetectorprocessor packages.
• No builder-config.yaml processors: wiring.
• No submodule tag bump (module/v0.2.0).
• No M19 cross-signal join test relocation (that comes with the
  processor packages).

All four land in PR-I.2b.

Test plan

• GOWORK=off go build ./... clean in root.
• GOWORK=off go build ./... clean in module/.
• cd module && go test ./pkg/patterns/... ./pkg/replay/... green
  (the rubric L405/L406/L414 gates pass at the new location).
• cd module && go test -race -count=1 ./pkg/patterns/... ./pkg/replay/... green (the corrected chaos.yml invocation).
• GOWORK=off go test ./... green in root (12 packages —
  internal/synthesis was the only consumer left in root
  post-PR-K).
• make check green (fmt, tidy-check, lint, vet, mod-verify).
• make verify green (adds license, fixtures, build-tags, RCE
  gate, register-lint, actionlint, zizmor, doc-check,
  no-autoupdate).
• module/go.mod collector pins still at v0.110.0 (MVS floor
  preserved post-tidy).
• No root-module Go file imports the moved packages (verified via
  grep); no tools/ or bench/ coupling.

PR-I.2a (relocate): pattern library + replay runner moved from
`internal/synthesis/{patterns,replay}/` to `module/pkg/{patterns,replay}/`
to colocate the moat surface with the Go submodule that will host the
OTel processors in PR-I.2b. Mechanical reshuffle only — no behavior
change, no API surface change, no public release artifact. Module pins
on `module/` stay at the OCB-distribution baseline (collector
v0.110.0). Partial close of #221; PR-I.2b (net-new processors +
builder-config wiring + `module/v0.2.0` tag) remains open.

Refs #221.