What this PR does / why we need it:

PR adjusts the Clickhouse offline store to be thread safe. Issue with the current implementation results in the issue as follows:

raise ProgrammingError('Attempt to execute concurrent queries within the same session.'

if you bombard it - feast serve - with concurrent requests, the under the hood threadpool is more than 1 thread by default (arbitrary number decided by uvicorn).

as per Clickhouse docs here:

Because each query or insert executed maintains state in its own QueryContext or InsertContext object, respectively, these helper objects are not thread-safe, and they should not be shared between multiple processing streams.

Solution - using threading.local() we can initialize a per-thread container for variables, to make sure we re-use the client on subsequent calls. Can we just drop the caching? Not really, unless the store is heavily refactored.

Which issue(s) this PR fixes:

No issue raised yet.

Misc