feat: add BUILD_TEMPLATE BDD opcode

[stobrien89]

What behavior changes?

Eligible string templates now compile to a fused BUILD_TEMPLATE instruction instead of multiple load/property instructions followed by RESOLVE_TEMPLATE.

Supported segments include literals, registers, and direct register properties. Complex expressions retain the existing fallback. Endpoint resolution results remain unchanged.

Generated bytecode now uses version 2, while version 1 bytecode remains readable.

Why is this change needed?

Template resolution currently requires an interpreter dispatch for each segment plus a final dispatch to concatenate them.

BUILD_TEMPLATE encodes eligible segments inline and resolves them in one dispatch, reducing interpreter and stack overhead while preserving existing allocation behavior.

How was this validated?

• Added compiler tests for registers, properties, fallback expressions, and endpoint host/path templates.
• Added evaluator, walker, malformed-bytecode, disassembler, and version compatibility tests.
• Added TemplateResolutionBenchmark.
• ./gradlew :rulesengine:check passes with all 257 tests.
• On x86-64 (m7i) with Corretto 25, JMH measured 50 samples per case across three to nine segments:
  • 3 segments: 2.5% lower register latency and 6.8% lower register-property latency.
  • 4 segments: 14.8% lower register latency and 5.4% lower register-property latency.
  • 5 segments: 14.9% lower register latency and 16.8% lower register-property latency.
  • 6 segments: 9.0% lower register latency and 13.0% lower register-property latency.
  • 7 segments: 10.1% lower register latency and 12.0% lower register-property latency.
  • 8 segments: 12.2% lower register latency and 15.8% lower register-property latency.
  • 9 segments: 12.4% lower register latency and 17.2% lower register-property latency.
• Allocations were unchanged across every benchmark case.

What should reviewers focus on?

• BytecodeCompiler: template eligibility and inline encoding.
• BytecodeEvaluator: fused template evaluation.
• BytecodeWalker: variable-length instruction validation.
• BytecodeDisassembler: symbolic template output.
• Opcodes and TemplateSegmentType: opcode and package-private segment tags.
• Bytecode: rolling version change from 1 to 2.
• TemplateResolutionBenchmark: comparison with RESOLVE_TEMPLATE.

Additional Links

• Prior art: the BUILD_TEMPLATE implementation in smithy-php.

---

By submitting this pull request, I confirm that my contribution is made under the terms of the Apache 2.0 license.