feat: update disk cache in another thread to avoid blocking normal query process

[jiacai2050]

Rationale

When there is a cache miss in disk cache, it will

1. Fetch data from remote
2. Insert data to cache, which will incur disk IO
3. Return the data for query.

We can move the second step to another thread to avoid it blocking the normal query process.

Detailed Changes

• Make write disk nonblocking
• Block on test explicitly, otherwise it will throw errors below

Cannot drop a runtime in a context where blocking is not allowed. This happens when a runtime is dropped from within an asynchronous context.

Test Plan

CI

[ShiKaiWi]

LGTM