[rahadbhuiya]

Summary

Closes #23

Introduces a configurable floating-point precision architecture across aegis-filter. By default, the crate retains maximum double-precision f64 headroom for aerospace and desktop environments. Enabling the opt-in cargo feature single-precision switches core matrix arithmetic, filter states, covariances, and FFI bindings to single-precision f32.

This eliminates 5x–15x CPU software emulation penalties on ARM Cortex-M4 and Cortex-M33 flight controllers equipped only with single-precision hardware FPUs.

Key Changes

1. Precision Abstraction (src/common.rs):
   • Added Real type alias (f64 by default, f32 under feature = "single-precision").
   • Extended FloatOps trait implementation to f32, including fast Newton-Raphson square root for #![no_std] targets.
2. Core Modules Refactored to Real:
   • src/linear.rs: ScalarKalmanFilter and LinearKalmanFilter<S, M>
   • src/robust.rs: HInfinityFilter1D and AdaptiveFilter1D
   • src/unscented.rs: UnscentedKalmanFilter2D
   • src/eskf.rs: ErrorStateKalmanFilter, Vector3, and Quaternion
   • src/fusion.rs: MultiSensorFusion1D and SensorSource
   • src/stats.rs: WindowedStats<N>
   • src/ffi.rs: C-FFI wrapper structs and functions
3. C-FFI Header (include/aegis_filter.h):
   • Introduced aegis_real typedef (defaults to double, switches to float when AEGIS_SINGLE_PRECISION is defined).
4. CI & Verification (.github/workflows/ci.yml):
   • Added bare-metal compilation check for thumbv7em-none-eabihf with --features single-precision.
   • Added test suite validation with --features single-precision.

Verification

• CI workflow runs both double-precision and single-precision bare-metal checks and unit tests.