Fix segmentation fault due to improperly implemented threading

[jbboehr]

This PR is intended to fix #8 by re-implementing the execution of callbacks passed to libzookeeper via the method used by the pcntl extension. Previously, callbacks would be executed in the thread started by zookeeper. Now callbacks will be executed in the original PHP thread either when calling zookeeper_dispatch(), by using declare(ticks=1), or, in PHP 7.1 or greater, automatically using VM interrupts.

[coveralls]

Coverage decreased (-19.7%) to 58.215% when pulling df26ce0 on thread-cb-issue-2 into d400640 on master.

[Timandes]

👍

[coveralls]

Coverage increased (+6.2%) to 84.07% when pulling 2649dd0 on thread-cb-issue-2 into d400640 on master.

[jbboehr]

Looks like the oneshot parameter is not currently working, and the internal callbacks may no longer be needed - requires some further review/work.

[Timandes]

If we are using NTS version of PHP, is this ZK_G safe enough to access the global variable head N tail?

[jbboehr]

I not sure NTS vs ZTS matters too much here, I believe ZTS is mainly required for environments where PHP is being run within many threads in the same process. [1]

I was considering adding a mutex to this block, but as there were none in pcntl, and portability may be an issue, I did not.

The pcntl extension may be a bit different, I guess signals probably use software interrupts instead of threads.

In any case, the conservative option would be to add a mutex here.

[coveralls]

Coverage increased (+5.7%) to 83.623% when pulling 6065f75 on thread-cb-issue-2 into d400640 on master.

[coveralls]

Coverage increased (+5.7%) to 83.623% when pulling 6065f75 on thread-cb-issue-2 into d400640 on master.