Fix read buffer compaction in stream filter flush

[crystarm]

php_stream_filter_flush() compacts unread data before appending buckets produced by a read filter.

The source and destination ranges may overlap, making the use of memcpy() undefined behavior. Additionally, readpos was reset before it was subtracted from writepos, so the buffer size was not adjusted and stale data could remain visible.

Use memmove() and adjust writepos before resetting readpos, matching the existing buffer compaction logic in php_stream_fill_read_buffer().

The issue was detected by static analysis: BUFFER_OVERLAP filter.c:[458:4].log

[crystarm]

Small clarification: the allocated buffer size (readbuflen) was not affected. Rather, resetting readpos before the subtraction left writepos unchanged, which could expose stale or duplicate bytes from the read buffer.

[LamentXU123]

Could you add this test please?

--TEST--
stream_filter_remove() compacts unread data before appending flushed data
--FILE--
<?php
class ClosingSuffixFilter extends php_user_filter
{
    public function filter($in, $out, &$consumed, $closing): int
    {
        while ($bucket = stream_bucket_make_writeable($in)) {
            $consumed += $bucket->datalen;
            stream_bucket_append($out, $bucket);
        }
        if ($closing) {
            stream_bucket_append($out, stream_bucket_new($this->stream, 'END'));
        }
        return PSFS_PASS_ON;
    }
}
stream_filter_register('closing-suffix', ClosingSuffixFilter::class);
$stream = fopen('php://memory', 'w+');
fwrite($stream, 'abcdef');
rewind($stream);
$filter = stream_filter_append($stream, 'closing-suffix', STREAM_FILTER_READ);
var_dump(fread($stream, 2));
var_dump(stream_filter_remove($filter));
var_dump(stream_get_contents($stream));
?>
--EXPECT--
string(2) "ab"
bool(true)
string(7) "cdefEND"

Also, could you please rebase to 8.4 instead of master?

[crystarm]

@LamentXU123
Addresed!! ദ്ദി(˵ •̀ ᴗ - ˵ ) ✧

[crystarm]

@LamentXU123
Maybe I should also rebase this PR onto 8.4?

[LamentXU123]

Not sure. The phar one is controversial and we should wait for other's agreement :) I think people will look back to it before the hard feature freeze.

[LamentXU123]

@crystarm Thank you so much! Note that I added a NEWS entry for you so what I merged is different to what you've wrote here.