Avoid allocations when getting content of DOM nodes if possible

[nielsdos]

Everytime text content or node value or something alike is requested, it currently allocates memory twice: once in getting the data from libxml2, and then allocating a zend_string to copy the data into.
We can very often skip the first allocation. Let's do that because it's actually quite common to get the textContent for example from a node when processing documents.

Here's a microbenchmark with time measurements (only for textContent, but for nodeValue etc the result is similar).
There's no noticeable slowdown because of the extra checks for the cases where we cannot avoid an allocation.
Based on these results I'd say it's worth it:

<?php

$doc = new DOMDocument;
$doc->loadXML('<?xml version="1.0"?><div/>');
$doc->documentElement->append(str_repeat('hello world', 10));
$el = $doc->documentElement->firstChild;

for ($i = 0; $i < 1000000; $i++) {
    $el->textContent;
}

Bench results:

Benchmark 1: ./sapi/cli/php dom.php
  Time (mean ± σ):      43.1 ms ±   2.2 ms    [User: 40.4 ms, System: 2.6 ms]
  Range (min … max):    40.3 ms …  51.5 ms    62 runs
 
Benchmark 2: ./sapi/cli/php_old dom.php
  Time (mean ± σ):     107.6 ms ±   1.6 ms    [User: 104.5 ms, System: 2.9 ms]
  Range (min … max):   105.4 ms … 112.1 ms    27 runs
 
Summary
  ./sapi/cli/php dom.php ran
    2.50 ± 0.13 times faster than ./sapi/cli/php_old dom.php

[Girgias]

LGTM

[Girgias]

Complete side note, changing the return type from various functions that only return SUCCESS and FAILURE from int to zend_result might be a good idea as a follow-up.