Connection failure leaves connection in an unusable state

[KPull]

When a query eventually made its way into the ExecuteContext, if establishing the connection raised an error within the __aenter__ section, the connection was not being cleared up. Subsequent queries to the same connection (in a pool) would state that the "previous query should be finished first."

In this pull request, we attempt to disconnect the connection (essentially doing the same thing __aexit__ does), marking it as stale within its associated pool, if the connection fails to be establish.

Thanks to @borgmatthew as well who helped with identifying the issue.

[itssimon]

This could potentially fix #93 too.

@long2ice mind taking a look at this?

[long2ice]

Thanks!

[itssimon]

Thanks for merging this! If you get a chance, would you mind creating a new release so we can benefit from this fix?

[SaneBow]

Nicely done! When can we have a new release available with this fix?

[tiejunhu]

Also expecting a new release for this fix