perf: optimize schema and serde hot paths

[jterapin]

Summary

Optimizes schema, serde, and client hot paths on top of the merged schema-extension foundation.

• Flattens and independently resolves schema, JSON, XML, and HTTP extension metadata.
• Reuses frozen JSON, XML, and CBOR codec workers across calls.
• Keeps XML build state local so worker reuse remains thread-safe.
• Reduces JSON, XML, and CBOR encoding and decoding allocations.
• Simplifies shape dispatch and removes redundant metadata lookups.
• Reduces client request preparation, conversion, validation, host-prefix, and response-status overhead.

Each commit isolates one optimization or correctness follow-up so the changes can be reviewed incrementally.

Review guide

Recommended review order:

1. Schema and extension metadata

   • Flat, namespaced metadata keys replace grouped hashes.
   • Boolean and nullable values use fetch_metadata where ||= would not memoize them.
   • Runtime shape dispatch uses shape classes instead of cached numeric tags.

2. Serde codecs

   • JSON, XML, and CBOR reuse frozen workers owned by each codec.
   • Workers retain configuration only; per-call state remains local.
   • XML passes its DocBuilder through each build rather than storing it on the reusable worker.

3. Allocation reductions

   • CBOR avoids temporary byte strings and repeated header decoding.
   • JSON and XML builders reduce intermediate objects and repeated lookups.
   • Client paths reuse resolved targets and cached metadata.

4. Safety and compatibility

   • String-based trait lookups remain in extension and serde paths for dynamic-client support.
   • Generic runtime union and float dispatch remains.
   • Sequential and concurrent codec reuse is covered by tests.
   • RBS signatures match the updated runtime APIs.

Non-goals

• No protocol wire-format behavior changes.
• No code-generated serde redesign.
• No removal of runtime union or float dispatch.
• No changes to timestamp precision.

Compatibility

The corresponding AWS Query and REST schema-extension changes have merged into aws-sdk-ruby-staging version-4 through PRs #76, #77, #78, #80, and #81.

Validation

• All GitHub CI checks pass.
• smithy-schema: 137 examples, 0 failures.
• smithy-json: 69 examples, 0 failures.
• smithy-xml: 58 examples, 0 failures.
• smithy-cbor: 268 examples, 0 failures.
• smithy-client: 889 examples, 0 failures, 2 pre-existing pending.
• Schema, JSON, XML, CBOR, and client RBS validation passed.
• Full RuboCop: 366 files, no offenses.
• Full staging specs and generated protocol suites passed during stack validation.

[richardwang1124]

Could using the same builder each time build is called cause multithreading issues? Previously each time build was called a new Builder was created.

[jterapin]

Good callout. I ended up making some changes so that we don't use any per-call state from now on. When we switch protocols, the codecs are re-init.