sql: support pgwire query cancellation

[rafiss]

Postgres docs describe how BackendKeyData is initialized by the DB on startup and sent to the client so it can be used for cancellation requests later.

https://www.postgresql.org/docs/12/protocol-flow.html

BackendKeyData

    This message provides secret-key data that the frontend must save if it wants to be able to issue cancel requests later. The frontend should not respond to this message, but should continue listening for a ReadyForQuery message.

CockroachDB has a separate cancel query syntax. For increased compatibility, we should support the Postgres version too.

There have been other attempts at doing this in the past, but they had some issues.

See #13009 #13191 #17003 #34520

Epic CRDB-7644

[awoods187]

@solongordon if this is relatively easy i'd be interested in picking it up because its used in part of the solution for #52240 and seems generally useful behavior for our drivers to posses.

[rafiss]

See #34520 for the latest attempt at implementing this. Eng has not really reached a consensus on how to do this. But I'm still hopeful!

Knowing that there's a product motivation for working on it could make it easier to drive consensus.

Also, I believe that we collect telemetry on how often a Cancel message is received. Not sure where to find it, but if anyone knows, I'd be curious to see the stats.

[awoods187]

It's the top unimplemented feature: https://cockroachlabs.looker.com/looks/47

We should prioritize this particularly if we have some prior art to draw upon.