docs: track making the generated Rust crate no_std

[mario4tier]

Tracking PR (don't forget) — do not merge as-is

Complete plan kept in this description (the in-repo docs/plans/*.md will be cleaned up once implemented).

Captures a deferred follow-on: migrating the generated Rust crate to #![no_std]. Filed so it isn't lost; it is not a blocker for the Rust abstract/introspection layer work (that module is already no_std-clean by design).

Assumption to validate

Crate-wide no_std is strictly better and feasible. Benefits: runs on any target with (eventually without) an allocator — WASM, embedded — and enforces the same zero-/bounded-allocation discipline the C reference already has with stack temp buffers.

Current state

• The generated crate ta_codegen_output/rust/ is std-only (declares no_std nowhere).
• The only heap blocker is runtime-sized Vec<f64> temp/scratch buffers in 17 of 163 indicator files. No String/format! anywhere in indicator code.
• The forthcoming abstract/introspection module is no_std-clean by design (&'static/const tables, enums, slices), so it adds no debt.

The 17 files: accbands, adxr, apo, atr, bbands, dema, ma, macd, macdext, mavp, natr, ppo, stoch, stochf, stochrsi, tema, trix

Example (atr.rs): range-sized scratch buffer mirroring the C reference's stack temp buffers.

Approach (the fix lives in the GENERATOR — tools/ta_codegen/src/backends/rust_lang.rs + scaffolding in main.rs — never hand-edit ta_codegen_output/)

• A — extern crate alloc (minimal, low risk, recommended first): emit #![no_std] + extern crate alloc; and switch temp buffers to alloc::vec::Vec. Runs on any allocator target incl. WASM; no algorithm/signature changes.
• B — true zero-alloc (higher fidelity to C, more work): replace runtime-sized Vec scratch with caller-provided/bounded/stack buffers, removing the allocator requirement. Touches the 17 functions and possibly the public surface. Defer unless a no-allocator target is required.

Recommended: A first, revisit B only if an allocator-free target becomes a real need.

Acceptance criteria

• Crate builds with #![no_std] (e.g. a no_std smoke/CI target)
• ta_regtest --codegen --language=rust still 161/161 — no behavior change
• No hot-path (per-bar compute) perf regression vs the current std build
• The abstract/introspection module remains no_std-clean
• Change lives in the generator + scaffolding; ta_codegen_output/ is regenerated, never hand-edited

🤖 Generated with Claude Code