[bench] Parser micro-benches (Turtle, JSON-LD, SPARQL)

[aaj3f]

Context

Follow-up bench from PR #1228 review (coverage gap #8). Turtle, JSON-LD, and SPARQL parser hot paths are exercised through import_bulk and transact_commit end-to-end, but never isolated. A regression in any of these parsers would surface here only as a small percentage of query_hot_bsbm setup or import_bulk time, where it's easy to mistake for noise.

Scope

Add fluree-db-query/benches/parser_micro.rs (or split into three files if that's cleaner — parser_turtle.rs, parser_jsonld.rs, parser_sparql.rs):

• Turtle parser: parse a gen::people-generated Turtle blob at Tiny/Small scales. Measure throughput in triples/sec (parse-only — no Fluree state).
• JSON-LD parser: parse a gen::people JSON-LD @graph document. Same metric.
• SPARQL parser: parse the BSBM Q3/Q5/Q9 strings from query_hot_bsbm.rs plus a few longer-shape queries (UNION + OPTIONAL + property paths). Measure throughput in queries/sec.

Each:

• Use the chassis pattern. current_profile(), current_scale(), init_tracing_for_bench(). No tokio runtime needed for pure parsing — bench_runtime() is unnecessary here.
• No Fluree state. This is parser-only, no need to build a Fluree.
• Register budgets in regression-budget.json. Tighter budgets here are appropriate since parsing is deterministic and CPU-bound.

Acceptance

• Bench(es) compile and run --test green at tiny.
• regression-budget.json has entries.
• BENCHMARKING.md's "Current benches" table grows one or three rows depending on the file split.
• Bench-gate CI job picks them up.

References

• PR Workspace-Wide Benchmarking Foundation #1228 review: coverage gap Fix browser cljs build #8.
• Parser locations: fluree-graph-turtle, fluree-graph-json-ld, fluree-db-sparql/src/parse/.

Out of scope

This is parse-only. Round-trip serialization correctness is owned by existing tests; this bench isolates parser perf.