Possible numerical underflow during quantization of log sampled UniformFloat

[NeoChaos12]

I noticed that in the case of UniformFloatHyperparameter types that have log-sampling enabled, there is a possible numerical underflow issue with using a quantization factor. To illustrate this, consider the following minimum working example:

import ConfigSpace as CS
p = CS.UniformFloatHyperparameter("param", 0.0001, 1.0, log=True, q=0.0001)

:1: RuntimeWarning: invalid value encountered in log

space = CS.ConfigurationSpace("sp1")
space.add_hyperparameter(p)

Traceback (most recent call last):
File "", line 1, in
File "ConfigSpace/configuration_space.pyx", line 186, in ConfigSpace.configuration_space.ConfigurationSpace.add_hyperparameter
File "ConfigSpace/configuration_space.pyx", line 1062, in ConfigSpace.configuration_space.ConfigurationSpace._check_default_configuration
File "ConfigSpace/configuration_space.pyx", line 1427, in ConfigSpace.configuration_space.Configuration.init
File "ConfigSpace/configuration_space.pyx", line 1457, in ConfigSpace.configuration_space.Configuration.is_valid_configuration
File "ConfigSpace/c_util.pyx", line 38, in ConfigSpace.c_util.check_configuration
File "ConfigSpace/c_util.pyx", line 103, in ConfigSpace.c_util.check_configuration
ValueError: Active hyperparameter 'param' not specified!

More cases that fail:

p = CS.UniformFloatHyperparameter("param", 0.0001, 1.0, log=True, q=0.00001)
:1: RuntimeWarning: invalid value encountered in log

p = CS.UniformFloatHyperparameter("param", 0.00001, 1.0, log=True, q=0.00001)
:1: RuntimeWarning: invalid value encountered in log

This persists even when the upper bound is smaller, such as here:

p = CS.UniformFloatHyperparameter("param", 0.0001, 0.01, log=True, q=0.0001)
:1: RuntimeWarning: invalid value encountered in log

Reducing the lower bound, however, fixes this. That is, this works fine:

p = CS.UniformFloatHyperparameter("param", 0.001, 1.0, log=True, q=0.0001) # but lower=0.0001 does not!

and removing the quantization factor works fine as well, even for a very small lower bound:

p = CS.UniformFloatHyperparameter("param", 0.00001, 1.0, log=True)

I haven't tested other numerical hyperparameters for similar cases, but it is definitely possible that they will run into similar issues.

My environment information:

OS: Ubuntu 18.04
Python: 3.7
ConfigSpace version: 0.4.19
NumPy version: 1.21.0