Add incremental parse API and extract parser engines

[default-anton]

What Changed

JsonCompleter.parse is the new fast path for callers that want Ruby values from partial JSON streams.

On the benchmark in this PR, it is 8.93x faster at p50 than complete + JSON.parse and reduces allocated objects 4.66x (20419154 -> 4381473).

.parse is the primary streaming path: it returns the current Ruby value directly from partial input while maintaining parser state between calls, so each update processes only new data. .complete remains for callers that specifically need completed JSON text.

The implementation is split into:

• ParserEngine for incremental .parse
• CompletionEngine for .complete
• Scanners for shared token scanning

The PR also separates parse/completion specs, adds a benchmark spec, and updates the README to position .parse as the primary streaming API.

Validation

• bundle exec rubocop
• bundle exec rspec
• JSON_COMPLETER_BENCHMARK=1 bundle exec rspec spec/parse_benchmark_spec.rb

Benchmark

Command: for i in $(seq 1 10); do JSON_COMPLETER_BENCHMARK=1 bundle exec rspec spec/parse_benchmark_spec.rb; done

Defaults: iterations=50, chunk_size=8 (roughly 2 streamed tokens per update).

Run Parameters

Payload bytes	Prefixes	Iterations	Chunk size	Samples
77690	9712	50	8	10

Runtime Summary

Metric	parse	complete + JSON.parse
Min total	1.9712s	17.7352s
P50 total	2.0006s	17.8693s
P90 total	2.0251s	18.1113s
P95 total	2.0525s	18.1960s
Max total	2.0799s	18.2806s
Mean total	2.0070s	17.9238s

Allocation and Memory

Strategy	Allocated objects	Heap growth bytes	Heap growth pages	Retained objects	Retained bytes
parse	4381473	0	0	8	448
complete + JSON.parse	20419154	196608	3	1	160

Comparison

Metric	Value
Speedup (min)	8.68x
Speedup (p50)	8.93x
Speedup (p90)	9.10x
Speedup (p95)	9.13x
Speedup (max)	9.16x
Speedup (mean)	8.93x
Object allocation reduction	4.66x