limits_concurrency on_conflict: :discard looking only into running jobs and not blocked jobs

[MRigal]

In the uncommon case that (because of whatever error, possibly ProcessPrunedError), a job is not running anymore but the Semaphore hasn't expired yet, SolidQueue is piling up new jobs in "Blocked", thus not fulfilling the on_conflict: discard promise.

I see two possible solutions:

1. Consider checking if a semaphore is blocking for the key
2. Check if a job is "running" not only by looking if there is a job in running or ready but also possibly in blocked

It's slightly an edge case, but if the cron is set up to a quite high frequency, it can pile up quite a bit in the few hours the semaphore hasn't expired.

NB: this happened on SolidQueue 1.3.1. We want to update soon and I have not seen in the changelog what could have fixed it "en passant"

[MRigal]

We actually found a flaw in how we trigger the job and that was probably the clue. I will try to reproduce once we have a fix, not sure when. You may also consider this issue as closed/irrelevant if you're sure or have a test about it