fix(rust): propagate math errors instead of masking with unwrap_or(ZERO)

[johngrantuk]

Reported issue:

.unwrap_or(U256::ZERO) converts failure into a plausible-looking value

Root cause behind most of finding 1, and worth a crate-wide pass.

The crate uses this fallback in 46 places — 16 in reclamm_math.rs, 15 in reclammv2_math.rs,
the rest spread across buffers, LBP, QuantAMM and hooks. Two problems compound:

• It is mostly dead code. mul_down_fixed and mul_up_fixed are declared -> Result<_, PoolError> but have no failing branch — they always return Ok. So the unwrap_or never fires
  and gives false reassurance.
• Where a helper can fail, zero is the worst possible substitute. Zero is a legal value that
  flows straight into the divisions and subtractions listed above, so a computation that could not
  be performed becomes a panic several frames later instead of an error at the point of failure.

Suggested fix: propagate with ? rather than substituting zero; and either give the infallible
helpers a non-Result signature or make them actually detect overflow.