Revert "fix(pool): always ping connection on release to see if it's still viable"

[dignifiedquire]

This reverts commit 0ed524d.

Unfortunately #1025 broke code that uses a SqlitePool, especially when using a max_connection count of 1.

[abonander]

I would prefer to find the underlying issue rather than reverting the fix as it probably just exposed another bug. SqliteConnection::ping() is a no-op which means it should be instantly released to the pool anyway. Could this possibly have been caused by #1080?

[dignifiedquire]

I bisected things, and this is the exact commit that breaks things

[abonander]

This commit may be triggering a bug I just spotted in DecrementSizeGuard::drop() because it changes the timing of release() by always spawning it as a background task: https://github.com/launchbadge/sqlx/blob/master/sqlx-core/src/pool/inner.rs#L402-L407

If the waker at the head of the queue has no strong refcounts left then this will return instead of attempting to wake the next one. With a pool of size 1 this means that any waiting tasks will stall until they time out.

[dignifiedquire]

This commit may be triggering a bug

I see that of course makes sense

[dignifiedquire]

fixed on master