Add support for per-field indices

[kirkrodrigues]

Request

Datasets typically have a set of fields that are frequently (or always) used for filtering. E.g., timestamp, log-level, hostname, etc. Although clp already performs coarse-grained indexing on certain fields like the records' timestamps, we need more generality in the form of:

• the ability to specify arbitrary fields that should be indexed.
• the ability to specify how those fields should be indexed (e.g., using different types of indices).
• the ability to perform this indexing on-demand, after archives have already been generated.

Additionally, this feature should be supported both within and across archives.

Ideally, these indices should provide a good balance between improved query performance while minimizing storage and management overhead.

Possible implementation

Nothing concrete as yet.