[heft-native] Opt-in Rust front end and plan-driven Node host

[Sean Larkin (TheLarkInn)]

Opt-in native heft. Nothing changes for existing users.

Stacked on #6097, so this diff is the Rust work only. Full diff vs main: main...thelarkinn/heft-native-rust

Before → after

heft 1.3.1 → native binary. Same setup as the v2 PR.

	Before	After	Speedup
heft --help, --version, <action> --help	413 ms	2.6 ms	159×
Build with only built-in tasks	425 ms	2.4 ms	177×
First task, JS plugins, cold	423 ms	85 ms	5.0×
First task, JS plugins, warm host (opt-in)	423 ms	26 ms	16×
Peak RSS, --help	~85 MiB	2.6 MiB	−97%

What it is

• apps/heft-native: a std-only Rust binary. No dependencies. No Node bindings. 838 KB.
• Rust handles the CLI, config, rigs, plugin resolution, schema validation, heft clean, and copy/delete/set-env builds.
• JS plugins run in a plan-driven Node host (apps/heft/src/host). Rust execs node with the plan.
• HEFT_WARM_HOST=1 keeps single-use hosts ready on a private Unix socket.

Not disruptive

• Not a Rush project. rush install and rush build work without cargo.
• @rushstack/heft defaults and API reports unchanged.

Try it

cd apps/heft-native && cargo build --release
cd ../../build-tests/heft-node-everything-test
../../apps/heft-native/target/release/heft --help

Limits

• Linux only. Windows compiles, untested.
• Interactive terminals, watch, --debug, --verbose and config errors defer to JS heft.
• Warm hosts cost ~14 MiB more each.
• New files carry no license header, per a no-comments rule. Needs a call.

---

Experiment stacked on this: #6099 — Round 4 assembly/SIMD squeeze (SIMD JSON scans, SHA-NI, syscall cuts, static link).