perf: tighten smithy serde hot paths (1/3)

[jterapin]

Part 1 of 3 splitting #358.

Summary

Consolidates JSON, XML, and CBOR serde optimizations plus the smithy-schema extension helpers they share.

Changes

• Cache list/map member and wire-name lookups on hot paths
• Reuse Builder and Parser instances within codecs
• Avoid coercion and helper indirection in union/structure handling
• Align JSON and CBOR parser lookups with schema metadata
• Honor protocol default timestamp behavior across serde
• Lazily buffer XML parser text chunks; cache flattened trait lookups
• Move sparse? and related helpers from ExtensionHelpers into Extension
• Focus serde specs on behavior

Benchmarks

Paired V3-vs-V4 serde runs on graviton and intel: ~26% faster than V3 at the median, zero regressions on either architecture.

Per-protocol delta vs V3 (graviton, all rows are wins):

protocol	worst	median	best
awsJson1_0	-9.4%	-22.0%	-79.9%
restJson1	-24.4%	-46.3%	-63.7%
restXml	-25.0%	-44.5%	-59.2%
awsQuery	-12.6%	-27.9%	-38.9%
rpcv2Cbor	-2.2%	-16.7%	-58.2%

Stack

• 1/3 (this PR): serde work (smithy-schema, smithy-json, smithy-xml, smithy-cbor)
• 2/3: refactor: consolidate smithy-client http binding and utility helpers (2/3) #360 smithy-client refactor + generator template changes
• 3/3: chore: fix rubocop violations (3/3) #361 rubocop cleanup

Validation

• bundle exec rspec per gem: smithy-schema (161), smithy-json (78), smithy-xml (67), smithy-cbor (265), smithy-client (897, 2 pending). All pass.