Bench improvements with config files

[marianotepper]

This PR introduces a few ease-of-use tools in jvector-examples. These tools are mainly related to codeless benchmarking through the use of config files for setting hyperparameters.

1. Bench now loads the list of available datasets from a YAML file. The list is provided in jvector-examples/yaml-configs/datasets.yml. The datasets in datasets.yml can be commented to run a subset, for example. Additionally, Bench now allows to specify a list of overquery values for each topK value.
2. It creates BenchYAML that allows to read config files with JVector hyperparameters in YAML format. For details on the format of these config files, see below.
3. It creates HelloVectorWorld with a single, clean, and simple example using ada002-100k. This example does not do a grid search over the hyperparameters, but it does try a few overquery values for topK=10 and topK=100. The corresponding config file is in jvector-examples/yaml-configs/ada002-100k.yml.

Here's an example YAML file showing what and how we can specify hyperparameters:

configVersion: 4 # do not change this number unless you know what you are doing

dataset: ada002-100k # needs to match one of the names in jvector-examples/yaml-configs/datasets.yml

construction:
  outDegree: [32, 48, 64, 96, 128]
  efConstruction: [60, 80, 100, 120, 160, 200, 400, 600, 800]
  neighborOverflow: [1.2f, 2.0f]
  addHierarchy: [No, Yes]
  compression:
    - type: None
    - type: PQ
      parameters:
        m: 192 # we can either specify the integer m or the integer mFactor. In this case, m will be set to the data dimensionality divided by mFactor
        # mFactor: 8
        k: 256 # optional parameter. By default, k=256
        centerData: No # optional parameter. By default, centerData=Yes if the dataset similarity function is Euclidean
        anisotropicThreshold: -1.0 # optional parameter. By default, anisotropicThreshold=-1 (i.e., no anisotropy)
    - type: PQ
      parameters: # here, the default values of k, centerData, and anisotropicThreshold are used
        mFactor: 2
  reranking:
    - FP
    - NVQ
  useSavedIndexIfExists: Yes

search:
  topKOverquery:
    # the value of topK followed by a list with the overquery rates we want to cover for that topK
    10: [1.0, 2.0, 5.0, 10.0]
    100: [1.0, 2.0]
  useSearchPruning: [No, Yes]
  compression:
    - type: None
    - type: PQ
      parameters:
        m: 192 # we can either specify the integer m or the integer mFactor.
        k: 256 # optional parameter. By default, k=256
        centerData: No # optional parameter. By default, centerData=Yes if the dataset similarity function is Euclidean
        anisotropicThreshold: -1.0 # optional parameter. By default, anisotropicThreshold=-1 (i.e., no anisotropy)

[tlwillke]

Looks good.