Fix leaking definitions on FFI::cdef()->new()

[iluuu1994]

Previously, FFI_G(symbols) and FFI_G(tags) were never cleaned up when calling new on an existing object. However, if cdef() is called without parameters these globals are NULL and might be created when new() creates new definitions. These would then be discarded without freeing them.

Furthermore, when calling new() on an existing object ffi->symbols and ffi->tags are reused for the operation. new() might add new definitions, leaking them into the next new() call. I'm not sure if this is desired behavior.

/cc @kocsismate

[kocsismate]

Nice, thank you! I also tried something similar but made a few mistakes along the way: fe14c4c#diff-74a5c960477012cea7bb9b723ea9b0e51279274a8703e145f45e7970316e2558R3712

[iluuu1994]

I'll merge this without the ext/ffi/tests/leaking_definition.phpt test, but it would be nice if @dstogov could verify whether that is a bug or intentional. For reference, in case of rebasing:

--TEST--
FFI::cdef()->new() shouldn't leak definitions into outer scope
--EXTENSIONS--
ffi
--XFAIL--
--FILE--
<?php
$ffi = \FFI::cdef('struct Example { uint32_t x; };');
var_dump($ffi->new('struct Example2 { uint32_t x; }'));
try {
    var_dump($ffi->new('struct Example2'));
} catch (\FFI\ParserException $e) {
    echo $e->getMessage(), "\n";
}
?>
--EXPECTF--
object(FFI\CData:struct Example2)#2 (1) {
  ["x"]=>
  int(0)
}
Incomplete struct "Example2" at line %d

- Incomplete struct "Example2" at line %d
+ object(FFI\CData:struct Example2)#2 (1) {
+   ["x"]=>
+   int(0)
+ }