feat: configurable arbitrary precision

[stefan-gorules]

Make arbitrary_precision an optional feature flag

This PR refactors serde_json's arbitrary_precision from being always-on to an opt-in Cargo feature flag across the core crates (zen-types, zen-expression, zen-engine). All bindings (Node.js, Python, C, UniFFI) explicitly enable the flag, preserving current behavior.

What changed:

The arbitrary_precision feature on serde_json was previously hardcoded in zen-types, zen-expression, and zen-engine. Now each crate declares its own arbitrary_precision feature that propagates through the dependency chain (zen-engine → zen-expression → zen-types → serde_json). Core crates default to arbitrary_precision off, while all binding crates opt in explicitly.

When the feature is enabled (current default for all bindings), number handling is unchanged — numbers round-trip through serde_json as strings and are parsed via Decimal::from_str_exact / from_scientific.

When disabled, serialization and deserialization fall back to native i64/u64/f64 conversions, and the NUMBER_TOKEN-based struct hack used by arbitrary_precision is compiled out via #[cfg] guards in the Variable and RcValue ser/de impls.

Additional changes:

• CI: replaced addnab/docker-run-action@v3 with stefan-gorules/docker-run-action@v1