[core:math/linalg] Add linalg.clamp_length(vector, max_length) -> clamped_vector

[Bazzagibbs]

Occasionally vectors need to be clamped to a maximum length while preserving smaller values, for example when processing joystick input from an unreliable source.

larger         := [2]f32 {1.0, 1.0}
larger_clamped := linalg.clamp_length(larger, 1.0)
// larger_clamped is approximately {0.707, 0.707}

smaller         := [2]f32 {0.2, 0}
smaller_clamped := linalg.clamp_length(smaller, 1.0)
// smaller_clamped == smaller

Divide-by-zero safety:

• a is explicitly checked to be > 0
• m2 is always 0+ because it's squared.
• a*a is always 0+ because it's squared.
• if m2 != 0, it's not divide-by-zero.
• if m2 == 0, then m2 <= a*a which returns the original vector.