APCng collect - do not throw exception if a metakey is missing

[ashandi]

These changes should fix the following issues:
#124
#126

The main idea of this PL is following: when we use APCng with ttl 0 we have no guaranty that our data will not be removed from the cache. So the case when we can't find some metakey in the cache is common and we shouldn't rise the exception that stops completely the metrics collection process.

The same solution is already implemented in the Redis Storage:

prometheus_client_php/src/Prometheus/Storage/Redis.php

Lines 407 to 409 in a09ea80

	if (!isset($raw['__meta'])) {
	continue;
	}

prometheus_client_php/src/Prometheus/Storage/Redis.php

Lines 503 to 507 in a09ea80

	$metaKey = $this->removePrefixFromKey($metaKeyWithPrefix);
	$rawSummary = $this->redis->get($metaKey);
	if ($rawSummary === false) {
	continue;
	}

prometheus_client_php/src/Prometheus/Storage/Redis.php

Lines 590 to 592 in a09ea80

	if (!isset($raw['__meta'])) {
	continue;
	}

prometheus_client_php/src/Prometheus/Storage/Redis.php

Lines 626 to 628 in a09ea80

	if (!isset($raw['__meta'])) {
	continue;
	}

No meta data - no statistics.

Another problem here is that the metadata stored only once for a metric. So if we loose metadata from the cache, the metric will continue collecting but we lost access to it. So I made calling of storeMetadata function always, not only on the first writing of the metric. This function checks inside that metadata is already exists in the cache and will not do anything if it is.

[ashandi]

According to this, this exception shouldn't be caught if the apc.shm_size setting has enough value.