Fix cycle detector issue with checking for blocked actors

[dipinhora]

As part of PR #2800, I broke the cycle detector due to incorrect
incremental logic for querying actors for if they're blocked. I
relied on the hashmap internal array/bucket index value instead
of keeping tracking of # of actors queried for if they're blocked
during the current function call.

This commit resolves the logic issue described above by keeping
a running count of actors queried during the current function
call.

This commit also implements @plprofetes suggestion to make the
threshold for # of actors to query for if they're blocked into
a dynamic value. Specifically, the threshold is now the larger
of the CD_MAX_CHECK_BLOCKED constant (which is 1000) or 10%
of the total number of actors currently known to exist by the
cycle detector.

Resolves #2975.

[dipinhora]

This was tested using @plprofetes' example program in #2975.

The change regarding the dynamic threshold for # of actors to query if they're blocked is somewhat controversial as it can result in message storms originating from the cycle detector. Let me know if that change should be reverted or separated out into its own PR.

[jemc]

Specifically, the threshold is now the larger
of the CD_MAX_CHECK_BLOCKED constant (which is 1000) or 10%
of the total number of actors currently known to exist by the
cycle detector.

Just wondering, does the name CD_MAX_CHECK_BLOCKED no longer make sense, then?

[dipinhora]

Possibly. Got any ideas for a better name?

[jemc]

I keep coming up with bad naming ideas involving the "minimum maximum", so maybe not. To avoid confusion for future readers, I'm just suggesting we add this comment to make sure the reader knows how the constant is intended to be used, so that they can tell it is working as intended, and not see it as a bug that we may have a limit that is larger than the "maximum".

[SeanTAllen]

@dipinhora can you add release notes for this?

[dipinhora]

release notes (possibly):

Fix issue #2975 where the cycle detector wasn't correctly reaping cycles of dead actors.
Change cycle detector to use the maximum of 1000 or 10% of all actors currently alive for how many actors to query when the cycle detector periodically checks to see which actors are currently blocked. This is based on a suggestion by @plprofetes to ensure that the cycle detector is effective even for applications with a large number (millions) of temporary actors.