fix: precise query id rollover semantic

[wyfo]

Allow managing rollover and uniqueness, as well as being faster to access.

[github-actions]

PR missing one of the required labels: {'enhancement', 'dependencies', 'breaking-change', 'internal', 'documentation', 'bug', 'new feature'}

[Mallets]

@wyfo what's the status of this PR?

[wyfo]

I've realized that just making the query id available right after a timeout is not appropriate, because some response may come after. A better solution in case of timeout would be to add another delay, for example 1min in which the query id cannot be reused. And to handle the case where a lot of queries time out, those pending ids could be be inserted in a dedicated set, to be reused only in case of available id shortage.
@Mallets What do you think about it?

[Mallets]

I think the proposed change introduces quite a bit of complexity. We could simply rollover on QueryId assuming the query is no longer alive, which is a realistic assumption given the number of queries required to be performed in parallel.

[wyfo]

I think it's a valid assumption, but it quite defeats the point of having varint encoding for request id.

[Mallets]

There might be some room for optimization indeed but I don't believe it's critical.

[wyfo]

I've updated the PR taking in account the recent discussion.