Optimizing emalloc and friends

Hi, the attached patch is the beginning of introducing emalloc_t and friends to the engine which passes TSRM information to emalloc. Or to be correct when the full work is done the _emalloc calls are replaced by _emalloc_t calls which no longer need to do TSRMLS_FETCH(). As you can see i haven't done everything but from what i measured so far i'd expect a performance increase of about 10%. Also when all is done there will be NO difference when in non ZTS mode. And since i left the old function names as inline funcs the API doesn't even change. However it is a good idea to change all the _emalloc calls to the new _emalloc_t calls as i've done with the streams stuff already. If we can find a consensus here i'll work that stuff out and apply it asap. regards marcus