Function normalize_excess 

pub fn normalize_excess(value: f64) -> f64

Normalize excess gain/loss to [0, 1] using precomputed tanh lookup table.

Tanh is mathematically natural for this purpose:

• Maps [0, ∞) → [0, 1)
• Zero input → zero output
• Monotonically increasing
• Smooth gradients for backpropagation

The scaling factor (5.0) is derived from the observation that typical ITH excess gains range from 0 to 20%, and we want this range to occupy most of the [0, 0.8] output space.

Issue #96 Task #197: Uses precomputed LUT in 0.1 steps [0, 5] range instead of exp() (50-100 CPU cycles → <1 CPU cycle).

Arguments

• value - Raw excess gain or loss (absolute value used)

Returns

Normalized value in [0, 1)