Implement typeahead search

[farshidz]

Change Summary

Implement typeahead search. (Phase1)

The typeahead API can:

• Index popular queries to suggest from
• Generate suggestions from partial or full user input

Suggestion are generated in two stages:

1. Retrieval: Match all indexed queries where at least one input word matches a prefix of at least one query word, within a configured edit distance.
2. Rank by a score function consisting of query popularity and the match BM25 score.

We will implement this feature in two phases, and this PR only covers phase 1.

Phase 1 (in this PR): Basic features

• Add a new schema for typeahead, during index creation time
• Indexing with query normalisation
• Prefix query with fuzziness support, tunable weight on bm25 score and popularity
• Support random number fields in a metadata map fields (for display purpose only, e.g. hit count, etc.)
• Timestamp field to track the last updated epoch time
• Add the schema to existing indexes at bootstrapping time.

Phase 2 (future work): Advanced features

• Support score modifiers based on the metadata field, so user can define ranking based on random fields or multiple fields.
• Support Vespa garbage collection mechanism to automatically delete queries based on a sliding window.. This is to facilitate any ETL pipeline.
• Support schema upgrade (just for typeahead schema) to support new typeahead features for existing indexes
• (TBD) Configurable/Smart normalisation?

Related Jira Ticket

https://s2search.atlassian.net/browse/MOSD-394

Checklist

• Tests have been added for changes
• Documentation has been updated
• Breaking changes are clearly identified
• Python client changes linked or N/A

For new field types:

• Tests cover score modifier usage of this new type
• Test indexes updated to cover the new type for all APIs (add docs, search, partial update, etc.)

[farshidz]

lgtm