Lums/sc 40408/complete index vamana (#230)

This PR implements the coding portion of sc-40408.

* Brought the implementation of `vamana_index` to use the new `index_group` and `index_metadata` classes (via CRTP).
  * Defined the group directory structure for the index to include an array for the feature vectors and the adjacency structure of the graph, stored in CSR format, using an array for the neighbor ids, for the distance to the neighbors, and a partitioning index.
    *  `feature_vectors`
    * `adjacency_scores`
    * `adjacency_ids`
    * `adjacency_row_index`
  * Metadata includes
    * datatypes for index, scores, and neighbor ids
    * `L`, `R`, `alpha_min`, `alpha_max`, and `medoid`
    * `num_edges_history` - representing the size of the graph in number of edges.  The number of vertices in the graph is the same as the number of vectors and is stored in `base_sizes`.

The following schemas are taken from the group index in `test_data/nano/vamana/vamana_test_index`
```
# adjacency_ids
ArraySchema(
  domain=Domain(*[
    Dim(name='rows', domain=(0, 923), tile=924, dtype='int32', filters=FilterList([ZstdFilter(level=-1), ])),
  ]),
  attrs=[
    Attr(name='values', dtype='uint64', var=False, nullable=False, enum_label=None),
  ],
  cell_order='col-major',
  tile_order='col-major',
  sparse=False,
)

# adjacency_row_index
ArraySchema(
  domain=Domain(*[
    Dim(name='rows', domain=(0, 231), tile=232, dtype='int32', filters=FilterList([ZstdFilter(level=-1), ])),
  ]),
  attrs=[
    Attr(name='values', dtype='uint64', var=False, nullable=False, enum_label=None),
  ],
  cell_order='col-major',
  tile_order='col-major',
  sparse=False,
)

# adjacency_scores
ArraySchema(
  domain=Domain(*[
    Dim(name='rows', domain=(0, 923), tile=924, dtype='int32', filters=FilterList([ZstdFilter(level=-1), ])),
  ]),
  attrs=[
    Attr(name='values', dtype='float32', var=False, nullable=False, enum_label=None),
  ],
  cell_order='col-major',
  tile_order='col-major',
  sparse=False,
)

# feature_vectors
ArraySchema(
  domain=Domain(*[
    Dim(name='rows', domain=(0, 127), tile=128, dtype='int32', filters=FilterList([ZstdFilter(level=-1), ])),
    Dim(name='cols', domain=(0, 230), tile=231, dtype='int32', filters=FilterList([ZstdFilter(level=-1), ])),
  ]),
  attrs=[
    Attr(name='values', dtype='float32', var=False, nullable=False, enum_label=None),
  ],
  cell_order='col-major',
  tile_order='col-major',
  sparse=False,
)
```

The following is a dump of the metadata for the index group:
```
### Array Metadata ###
- Key: adjacency_row_index_datatype
- Value: 10
- Type: DataType.UINT32

### Array Metadata ###
- Key: adjacency_row_index_type
- Value: uint64
- Type: DataType.STRING_UTF8

### Array Metadata ###
- Key: adjacency_scores_datatype
- Value: 2
- Type: DataType.UINT32

### Array Metadata ###
- Key: adjacency_scores_type
- Value: float32
- Type: DataType.STRING_UTF8

### Array Metadata ###
- Key: base_sizes
- Value: [0, 10000]
- Type: DataType.STRING_UTF8

### Array Metadata ###
- Key: dataset_type
- Value: vector_search
- Type: DataType.STRING_UTF8

### Array Metadata ###
- Key: dimension
- Value: 128
- Type: DataType.UINT32

### Array Metadata ###
- Key: dtype
- Value: float32
- Type: DataType.STRING_UTF8

### Array Metadata ###
- Key: feature_datatype
- Value: 2
- Type: DataType.UINT32

### Array Metadata ###
- Key: feature_type
- Value: float32
- Type: DataType.STRING_UTF8

### Array Metadata ###
- Key: id_datatype
- Value: 10
- Type: DataType.UINT32

### Array Metadata ###
- Key: id_type
- Value: uint64
- Type: DataType.STRING_UTF8

### Array Metadata ###
- Key: index_type
- Value: Vamana
- Type: DataType.STRING_UTF8

### Array Metadata ###
- Key: ingestion_timestamps
- Value: [0, 1704946748930]
- Type: DataType.STRING_UTF8

### Array Metadata ###
- Key: num_edges_history
- Value: [0, 40000]
- Type: DataType.STRING_UTF8

### Array Metadata ###
- Key: storage_version
- Value: 0.3
- Type: DataType.STRING_UTF8

### Array Metadata ###
- Key: temp_size
- Value: 0
- Type: DataType.UINT64

```

A future story / PR should Test and benchmark -- there seems to have been some bit rot
* Use CLI programs for `vamana_index` for benchmarking
  * The type-erased class would be useful for this
* More fully / formally document the array schemas and metadata

Author: lums658