patch: no_std support (libm, core::time, remove expect)

[InushaDeSilva]

No description provided.

[InushaDeSilva]

Hey! We're using VQF on an embedded flight computer (Horizon-Aerospace/Flight-Computer) running on an STM32H745 (Cortex-M7, 480 MHz) with Embassy async. Great crate — the filter quality is excellent and the rest-detection/bias-estimation is exactly what we needed for our BMI088 IMU fusion pipeline.

What this PR changes

The vqf crate currently requires std due to a few things that have straightforward no_std replacements:

std usage	no_std replacement
std::time::Duration	core::time::Duration (stable since Rust 1.25)
.cos(), .sin(), .sqrt(), .tan() (bare float methods)	libm::cosf, libm::sinf, libm::sqrtf, libm::tanf
.to_radians()	* (core::f32::consts::PI / 180.0)
.powi(2)	x * x
.expect() on try_inverse()	match + early return (avoids pulling in panic formatting)

Cargo.toml changes

• Added std feature (enabled by default) — existing users see zero changes
• nalgebra switched to default-features = false, features = ["libm"] when std is off
• Added libm = "0.2" dependency

Tested

• cargo build --target thumbv7em-none-eabi --no-default-features — builds clean ✅
• Default cargo build (with std) — unchanged behavior ✅
• Running on hardware at 100 Hz fusion rate with BMI088 accel+gyro, producing stable roll/pitch/yaw + bias-corrected angular rates

Happy to adjust anything if you'd prefer a different approach. Thanks for the great crate!

[oxkitsune]

Thanks! I'd prefer to keep using std math for the std version if possible.

[InushaDeSilva]

Hey, thanks for the feedback! Both points have been addressed:

1. std math for std version 0 Added a math.rs wrapper module that delegates to std float methods (.sqrt(), .sin(), .cos(), .tan()) when std is enabled, and falls back to libm only when building with no_std.

2. Silent try_inverse failure - The else branch now panics in debug builds via #[cfg(debug_assertions)] so bugs surface early during development. In release/no_std builds it gracefully skips the bias update, which avoids pulling in panic formatting machinery on embedded targets. In practice this matrix should always be invertible, so the fallback is just a safety net.

Also cleaned up clippy lints and formatting. Let me know if anything else needs adjusting!