Separating Index Parameter Configuration into its Own Class

[BlaiseMuhirwa]

At a high level, this PR is just adding a IndexParameterConfig class that allows us to configure parameters that can be re-used to create different index objects. The python interface looks like this:

import flatnav
from flatnav.index import create_index, ParameterConfig
import numpy as np

parameter_config = (
    ParameterConfig(dataset_size=10000)
    .with_index_params(
        {
            "max_edges_per_node": 16,
            "ef_construction": 200,
            "num_initializations": 100,
            "num_threads": 4,
            "index_name": "hnsw_16_200"
        }
    )
    .with_reordering(["gorder", "rcm"]) 
)


index = create_index(distance_type="l2", dim=100, parameter_config=parameter_config)
index.add(data=np.random.rand(10000, 100))


# ##########################################
# # second initialization via a MTX file
# ##########################################


outdegree_table = flatnav.util.load_from_mtx_file(filename=hnsw_base_layer_filename)
index.allocate_nodes(data=train_dataset).build_graph_links(outdegree_table=outdegree_table)

I validated that this change doesn't lead to a performance regression in terms of QPS/latency.

[vihan-lakshman]

Is this a stray comment?