Optimize memcache::valid()

New patch.

Fixes a variable mis-name.

Only records a wildcard clear if cid is passed in - this code should catch those anyway:

// Determine when the current bin was last flushed.
    $cache_flush = variable_get("cache_flush_$this->bin", 0);

    $cache_lifetime = variable_get('cache_lifetime', 0);
    $item_flushed_globally = $cache->created && $cache_flush && $cache_lifetime && ($cache->created < min($cache_flush, time() - $cache_lifetime));

Moved some things up the stack to remove some function calls, a bit less code this way as well I think.

Noticed a few things in PHP which could be optimized slightly for cases when there's a multiple get of hundreds or thousands of cids, but those are for another issue really.

Looked at this a bit more since valid() was still taking around 6% of the request or more according to xdebug. This is on a page which calls valid() around 1000 times where the bulk of the cids being validated are via a couple of calls to cache_get_multiple(), so it's unusual, but I don't think it'll do any harm to normal pages either.

What this does:

* Consolidates various variable_get() calls into a single static along with the results of a min().
* Moves things around a bit so the more likely checks for an item to be expired come first.
* Swaps time() for REQUEST_TIME

According to xdebug/webgrind this is the overall effect, numbers will be a bit skewed due to xdebug overhead for function calls but still looks like a valid improvement to me:
Patch from #2:
total request time: 1365ms
valid() self: 82ms, incl: 140ms

This patch:
total request time: 1293ms
valid() self: 25ms, incl: 26ms

minus one change that was silly..

Remove the static and put the variable preparation into __construct() which I was going to add until I noticed there already was one. This removes the nastiness of trying to avoid the variable_get() now, no performance penalty from previous patches.

Had a <= when there should've been a >=. With this, the patch passes all tests, except for one fail when cache_lifetime is set, which also fails in HEAD, that might be by design.

function wildcards() had two hunks setting the wildcard_timestamps variable, removed one of them, I think that was patch cruft from an earlier roll. I think this is pretty much there now.

There should be a limited number of bins that have wildcard clears. I'm not sure the wildcard logic itself can get a lot more optimized than it is, did have one idea but no code yet.