Un-revert #1114

[mpilquist]

I had to revert #1114 on main due to a test failure in PoolTest:

==> X tests.PoolTest.progress and safety with many fibers and reset failures  0.044s skunk.util.Pool$ResourceLeak:
🔥
🔥  ResourceLeak
🔥
🔥    Problem: A resource leak was detected during pool finalization.
🔥     Detail: Expected 10 active slot(s) and no deferrals, found 107 slots and 0
🔥             deferral(s).
🔥       Hint: The most common causes of resource leaks are (a) using a pool on a
🔥             fiber that was neither joined or canceled prior to pool finalization,
🔥             and (b) using `Resource.allocated` and failing to finalize allocated
🔥             resources prior to pool finalization.
🔥

skunk.util.Pool$ResourceLeak: A resource leak was detected during pool finalization.
    at skunk.util.Pool$.$anonfun$ofF$21(Pool.scala:186)
    at cats.Applicative.whenA(Applicative.scala:264)
    at cats.Applicative.whenA$(Applicative.scala:263)
    at void @ org.typelevel.keypool.KeyPool$.reap(KeyPool.scala:187)
    at *>$extension @ org.typelevel.keypool.internal.RequestSemaphore$.apply(RequestSemaphore.scala:79)
    at *>$extension @ org.typelevel.keypool.internal.RequestSemaphore$.apply(RequestSemaphore.scala:79)
    at get @ org.typelevel.otel4s.sdk.trace.processor.BatchSpanProcessor.org$typelevel$otel4s$sdk$trace$processor$BatchSpanProcessor$$exportAll(BatchSpanProcessor.scala:160)
    at flatMap @ skunk.util.Pool$.free$1(Pool.scala:178)
    at foreverM$extension @ org.typelevel.otel4s.sdk.trace.processor.BatchSpanProcessor.org$typelevel$otel4s$sdk$trace$processor$BatchSpanProcessor$$worker(BatchSpanProcessor.scala:100)
    at mapN @ org.http4s.ember.client.EmberConnection.isValid(EmberConnection.scala:50)
    at unsafeToFuture @ munit.CatsEffectSuite$$anonfun$1.applyOrElse(CatsEffectSuite.scala:82)

Revert was done in 932002727c0aacb117217c489614211d2728235. Let's un-revert that and determine what needs fixing.