Ensure actor cancel callbacks are cleaned up

[temas]

While testing a long running web application that we uncovered a slow building memory leak related to cancel callbacks. The application did very little tile work, but did very high rate source data updates on multiple layers.

The primary issue found was that priority 0 tasks would store a cancel callback that would never be cleared from the actor::cancelCallbacks array. In tracing the code it seemed that priority 0 could never be truly canceled, so we opted to remove the callback return. We still saw a few issues after this change so we also moved the cancel callback delete to the start of the actor::processTask function. The function appeared no longer cancellable at this point and this created a safe cleanup.

With these minor changes, the application has been run for many hours, across multiple test machines, with no memory increase.

coauthor: @nsatter

Possibly helps for #12400

[stepankuzmin]

Hi @temas and @nsatter,

Good catch, thanks for making a PR for this! It looks good to me, but it's worth @mourner to also take a look.

[CLAassistant]

All committers have signed the CLA.

[mourner]

Looks good to me, just waiting on the CLA agreement to merge this. Thanks for the contribution!

[temas]

Bump for you all, this should be good to go.