Re-promote "We stalled too much while trying to write" warning if it happens too much

In #5286 (moved), implemented in a790454, we demoted the message that we give when a hold_open_until_flushed connection gets timed out. Previously it was at INFO or NOTICE depending on a number of aspects of the connection; now it's unconditionally INFO.

We discussed on #5286 (moved) that we should maybe re-promote the severity to NOTICE again (or even WARN?) if it happens a lot. I'm opening this ticket in case we decide to do that.

(See the commit a790454 for info on how we used to decide on whether to log INFO or NOTICE.)