perf: add processor hot-path benchmarks

[devarismeroxa]

Summary

This repo had zero benchmarks despite Processor.Process and reference
resolution (internal/reference) sitting on the per-record hot path for
every processor invocation. Adds committed benchmarks so future changes
to these paths have a baseline to regress against.

• internal/reference/reference_bench_test.go:
  • BenchmarkNewResolver - one-time reference-string parse/validate cost
    (paid once, typically at Configure).
  • BenchmarkResolver_Resolve - per-record resolve + Get(), covering
    raw vs. structured payloads, top-level fields (.Key,
    .Payload.After), metadata fields (plain and bracket-indexed), and
    structured sub-fields (including nested).
  • BenchmarkResolver_ResolveAndSet - per-record resolve + Set(),
    the pattern used by field-set/rename-style processors.
• processor_func_bench_test.go:
  • BenchmarkProcessorFunc_Process_MetadataFieldSet and
    BenchmarkProcessorFunc_Process_StructuredFieldSet - a
    representative field-set processor built on ProcessorFunc (the
    SDK's only concrete, reusable Processor implementation), exercising
    a full 100-record batch Process call end to end.

The SDK itself ships no concrete field-set/rename processor (those live
in the main conduit repo), so the Process-level benchmark uses
ProcessorFunc wired to the same resolve-then-set pattern real
processors use — the closest thing to "a representative processor" that
this repo actually exposes.

No production code changed.

Numbers (Apple M3 Max, go test -bench . -benchmem -benchtime 2s ./...)

Reference resolution (per record, after NewResolver is done once):

Benchmark	ns/op	B/op	allocs/op
Resolve .Key (raw)	3.5	0	0
Resolve .Key (structured)	3.4	0	0
Resolve .Metadata.field (raw)	40.8	40	2
Resolve .Metadata.field (structured)	39.5	40	2
Resolve .Payload.After (raw)	5.4	0	0
Resolve .Payload.After (structured)	5.4	0	0
Resolve .Payload.After.status (structured)	31.6	48	1
Resolve .Payload.After.customer.name (structured, nested)	93.8	120	3
Resolve+Set .Metadata.field (raw)	31.4	24	1
Resolve+Set .Metadata.field (structured)	30.7	24	1
Resolve+Set .Payload.After.status (structured)	34.4	48	1
Resolve+Set .Payload.After.customer.name (structured, nested)	101.7	120	3

NewResolver (paid once per reference string, not per record): 210 ns –
720 ns, 3–11 allocs, depending on reference depth.

ProcessorFunc.Process on a 100-record batch:

Benchmark	ns/op (batch)	ns/op (per record)	B/op	allocs/op
MetadataFieldSet / Raw	11,296	~113	26,592	301
MetadataFieldSet / Structured	11,619	~116	26,592	301
StructuredFieldSet (nested) / Structured	18,262	~183	36,192	501

Test plan

• go build ./...
• go test ./... -count=1
• go test ./... -count=1 -race
• golangci-lint run ./... - 0 issues
• GOOS=wasip1 GOARCH=wasm go build ./... - WASM build unaffected
• go test -bench . -benchmem -benchtime 2s ./... - all benchmarks pass, numbers above

🤖 Generated with Claude Code

https://claude.ai/code/session_01Tapg9dLWXKMZJoL65R24vd