tech-debt: migrate JSON serialization to orjson for performance

[dgarros]

Need

The SDK is a foundational library on the hot path of transforms, generators,
checks, exports/imports, and every GraphQL round-trip — JSON encode/decode
throughput directly affects how fast SDK-based automation runs at scale.
Today the SDK relies on ujson (and stdlib json in places). orjson, backed
by Rust (serde_json), is substantially faster than both and stricter on
correctness, so adopting it would give a measurable, no-behaviour-change speedup
for users running large syncs and pipelines.

Context

• This is a deliberate follow-up to tech-debt: standardize JSON library usage (json vs ujson) across the SDK #1165 (JSON library standardization). That
  issue picks a single library across the codebase; this one proposes making
  that target orjson rather than settling on ujson. Sequencing/merging the
  two is up to whoever picks them up.
• Adding orjson is a new runtime dependency — flagged here because the repo
  asks to confirm dependency additions.
• Known migration constraints the implementer will hit (not a design, just
  gotchas): orjson.dumps() returns bytes rather than str, so str(...) /
  .encode() wrappers at call sites like playback.py:52 and utils.py:253
  change; and orjson has no sort_keys= kwarg (it uses
  option=orjson.OPT_SORT_KEYS), which dict_hash (utils.py:253) depends on.

References

• Related issue: tech-debt: standardize JSON library usage (json vs ujson) across the SDK #1165
• orjson: https://github.com/ijl/orjson