Uniform quantizer: small step causes amplitude saturation for large values

[yep627]

Hi, thanks for releasing DASPack — it’s a very useful tool.
I noticed a numerical behavior with Quantizer.Uniform(step=...) that I think is worth reporting.

Observed behavior
When the data amplitude becomes relatively large (e.g. > 20), using very small step values leads to severe distortion:
step = 1e-7: works correctly up to amplitudes ~80+
step = 1e-8: decoded values seem to saturate around ~20
step = 1e-9: decoded values only remain correct up to ~1
The compression runs without error, but the recovered signal is clearly incorrect in amplitude.

Interpretation
It appears that the Uniform quantizer effectively performs:
q = round(x / step)
and that the internally stored integer residuals are limited by a finite integer range (possibly int32).
As a result, when:
|x| / step ≳ 2^31
quantized values overflow or saturate, leading to silent amplitude clipping after decoding.
This explains why decreasing reduces the maximum recoverable signal amplitude.step

Suggestion
It might be helpful to:

• Document the implicit relationship between and maximum safe data amplitude, or step
• Warn / raise an error when exceeds the internal integer range, or max(|data|) / step
• Clarify whether a larger integer type could be used internally for Uniform quantization

I hope this is useful feedback.

Best regards