fix: Allow partial priors

[Sohambasu07]

Features and/or fixes:

• Apply a fix here which replaces self._meaningful_ixs, which causes the code to crash in the presence of Categoricals in the space or non-prior Parameters in between Paraneters with priors. Fix for issue Enhancing Flexibility in Priorband Usage: Allowing Partial Knowledge Without Mandatory Default Parameters #28
• Allow partial priors in REQUIRES_PRIOR test here

Code for Reproducibility:

import numpy as np

import neps
from neps import AskAndTell, algorithms
from neps.space.parsing import convert_to_space


def evaluate_pipeline(float1, float2, float3, float4, cat, float5, float6):
    return -float(
        np.sum([float1, float2, float3, float4, float5, float6])
    )

pipeline_space = {
    "float1": neps.Float(lower=0, upper=1, prior=1, prior_confidence="high"),
    "float2": neps.Float(lower=75, upper=1000, prior=85, prior_confidence="high"),
    "cat": neps.Categorical(choices=["a", "b", "c"]),
    "float3": neps.Float(lower=1, upper=100, is_fidelity=True),
    "float4": neps.Float(lower=4, upper=6),
    "float5": neps.Float(lower=0, upper=10, prior=5, prior_confidence="high"),
    "float6": neps.Float(lower=1, upper=3),
}

optimizer = AskAndTell(
    algorithms.PredefinedOptimizers["priorband"](
        space=convert_to_space(pipeline_space),
    )
)

def set_seed(seed: int):  # noqa: D103
    import random

    import numpy as np
    import torch

    np.random.seed(seed)
    random.seed(seed)
    torch.manual_seed(seed)

set_seed(0)
for i in range(200):
    print(f"Iteration {i + 1}")
    trial = optimizer.ask()
    x = trial.config
    y = evaluate_pipeline(**x)
    optimizer.tell(trial=trial, result=y)