feat: JMH benchmarks for DataDictionaryGenerator and CodeGeneratorJ - #69

Adds a self-contained JMH benchmark module to measure the generation pipeline cost targeted by PR #68 (cached JAXBContext, cached XPathExpression, buffered I/O).

New module: quickfixj-from-fix-orchestra-benchmarks

• DataDictionaryGeneratorBenchmark — measures DataDictionaryGenerator.generate(InputStream, File): covers the double JAXBContext.newInstance(), per-call XPath compile, and unbuffered FileWriter that PR perf: DataDictionaryGenerator — buffer output, cache JAXBContext and XPathExpression #68 eliminates
• CodeGeneratorJBenchmark — measures CodeGeneratorJ.generate(InputStream, File): covers the JAXBContext.newInstance() and per-file unbuffered writes
• pom.xml — JMH 1.37 deps + maven-shade-plugin; builds a self-contained benchmarks.jar via mvn package -DskipTests
• trade.xml — FIX Orchestra input resource (copied from generator test resources)

Usage

# Build
mvn -pl quickfixj-from-fix-orchestra-repository/quickfixj-from-fix-orchestra-benchmarks \
    -am package -DskipTests

# Full run (3 warmup + 5 measurement, 1 fork)
java -jar .../benchmarks.jar

# Quick smoke-check
java -jar .../benchmarks.jar -wi 1 -i 1 -f 1

To verify PR #68: run on master (baseline) then on copilot/improve-code-generation-performance and compare the Score column — lower ms/op confirms the improvement.