Inconsistent error handling in `suggest_int` and `suggest_float`

Expected behavior

More consistent error handling for suggest_int and suggest_float.

Reproducible examples (optional)

import optuna


# Suggest int. OK.
study = optuna.create_study()
trial = study.ask()
trial.suggest_int("x", 0, 10)
trial = study.ask() 
trial.suggest_int("x", 0, 10, step=2)  # OK.

# Suggest float. ValueError.
study = optuna.create_study()
trial = study.ask()
trial.suggest_float("x", 0, 10)
trial = study.ask()
trial.suggest_float("x", 0, 10, step=2)  # ValueError: Cannot set different distribution kind to the same parameter name.

# Note: Similar behavior reproducible without the second `study.ask()`s.

Additional context (optional)

The inconsistency stem from historical reasons and in particular inconsistent distribution types and a consistency check based on these types https://github.com/optuna/optuna/blob/master/optuna/distributions.py#L521-L522. For instance, to suggest floats with steps you previously had to use suggest_discrete_uniform, similarly for log suggest_loguniform, until suggest_float was introduced which covers both cases. Internally, suggest_float simply calls suggest_discrete_uniform, etc. and hence the associated distribution types become different depending on arguments. At the same time, this interface is not symmetric for suggest_int. There is no suggest_discrete_int or suggest_logint.

To properly fix this issue, we probably have to rethink what distribution types to provide and make sure that samplers and other components depending on these distributions are aware.

[himkt]

#3166 will fix the problem. (It has not been merged into the master branch yet)

[contramundum53]

This issue has already been solved.