Add Temporal support & incidental performance improvements

[cn-stephen]

Hi @AsyncBanana :)

This PR adds runtime support for Temporal types, which otherwise throw at runtime. These ship as fully supported in TypeScript 6.0 and are stage4 https://tc39.es/proposal-temporal/ and solid as heck. They're not going anywhere!

Incidental performance improvements and a couple of tweaks for readability as I was passing through.

Also, a little housekeeping (package-lock, .node-version, quality checks in CI) hope you don't mind.

I hope this is acceptable for your great library!

Benchmark results: master vs temporal

Applied benchmarks

microdiff (no cycles)

Benchmark	master	temporal	delta
PandaCSS Config	12.04 µs	11.51 µs	-4%
Scalar/OpenAPI	7.54 µs	7.40 µs	-2%
Vanilla Extract	7.00 µs	6.66 µs	-5%

microdiff (with cycles)

Benchmark	master	temporal	delta
PandaCSS Config	19.07 µs	13.84 µs	-27%
Scalar/OpenAPI	11.00 µs	8.78 µs	-20%
Vanilla Extract	9.04 µs	7.65 µs	-15%

Geometric mean (applied)

	master	temporal
microdiff (no cycles)	100%	100%
microdiff (with cycles)	144%	118%

Theoretical benchmarks

microdiff (no cycles)

Benchmark	master	temporal	delta
Large Object (300 props)	13.19 µs	13.41 µs	+2%
Large Diff (300 props)	11.33 µs	9.36 µs	-17%
Deeply Nested	123.12 µs	111.17 µs	-10%
Small Object	69.22 ns	67.39 ns	-3%

microdiff (with cycles)

Benchmark	master	temporal	delta
Large Object (300 props)	13.22 µs	13.53 µs	+2%
Large Diff (300 props)	11.33 µs	9.31 µs	-18%
Deeply Nested	160.75 µs	120.78 µs	-25%
Small Object	70.29 ns	67.00 ns	-5%

Geometric mean (theoretical)

	master	temporal
microdiff (no cycles)	100%	100%
microdiff (with cycles)	107%	102%

Summary

• No regressions in any benchmark
• Cycles-enabled path improved 15-27% across all applied benchmarks (Set replacing linear array scan + eliminating array copies)
• No-cycles path improved 2-5% on applied benchmarks (eliminated intermediate .map() allocation)
• Deeply nested with cycles improved 25% (160.75 µs -> 120.78 µs) — the biggest theoretical win
• Large diff improved ~17% in both modes (early-exit refactor)
• Temporal type support added with zero performance cost to existing workloads

[AsyncBanana]

Thanks! Could you split these changes into multiple pull requests for each feature/optimization?

[cn-stephen]

Split into:

#53
#54
#55

With 1 more to come from https://github.com/cn-stephen/microdiff/pull/1 but relies on the code changes in #55