Adjust HEBO to FloatDistribution with step

[nabenabe0928]

This PR adjusts HEBOSampler so that it can handle FloatDistribution with step properly.

Note

Originally, HEBOSampler falls back to sample_independent if there are any parameters with FloatDistribution and step.

Please use the following objective function for the verification!

def objective(trial: optuna.Trial) -> float:
    trial.suggest_float("f", -5, 5)
    trial.suggest_float("f-log", 1, 100, log=True)
    trial.suggest_float("f-step", -2, 7, step=0.5)
    trial.suggest_float("f-step-non-divisable", 2, 7, step=3)
    trial.suggest_int("i", -5, 5)
    trial.suggest_int("i-log", 1, 100, log=True)
    trial.suggest_int("i-step", -2, 6, step=2)
    trial.suggest_int("i-step-non-divisable", 2, 7, step=3)
    trial.suggest_categorical("c", ["a", "b", "c"])
    print(trial.params)
    return trial.number

[nabenabe0928]

@y0z
Could you check this PR?

[HideakiImamura]

LGTM.

[HideakiImamura]

@nabenabe0928 Could you resolve the conflict?

[nabenabe0928]

I noticed that I need to transform before hebo.observe as well.

[nabenabe0928]

I fixed so that HEBO now knows the exact indices for each parameter.