Fixes #211

I noticed there were significant pages missing from the web search.

After some investigation I found there were 2 circumstances where entries were missing.

In the first case, some entries were being excluded by the !$index['chunk'] check. I found these could easily be re-included by skipping this check for specific element values. You can see the list of these running the following query on the output/index.sqlite file generated by a phd run:

SELECT *
FROM ids
WHERE chunk = 0
	AND element IN (
		'refentry',
		'stream_wrapper',
		'phpdoc:classref',
		'phpdoc:exceptionref',
		'phpdoc:varentry'
	)

Additionally entries were excluded where they all had the same docbook_id because of the way the indexes array is indexed by this id. Because the existing search indexes rely on this id, it's not easy to resolve.

However, the current web search actually reworks these indexes to combine them anyway, and I'd created a pre-combined version of these indexes for #204 that does not rely on indexing by the docbook_id.

I've copied these changes to this PR and then modified them to pull the index list from the database without the deduplication. These will require further changes to php-web js/search,js and js/search-index.php - I've already made the changes to search.js for #204 (and the changes to search-index.php is just having it spit out the new -combined index file without any manipulation it currently does)

You can see the list of affected docbook_ids with the following query:

SELECT docbook_id
FROM ids
WHERE chunk = 1
	OR element IN (
		'refentry',
		'stream_wrapper',
		'phpdoc:classref',
		'phpdoc:exceptionref',
		'phpdoc:varentry'
	)
GROUP BY docbook_id
HAVING COUNT(*) > 1

Generally these are cases where there are both procedural and OOP interfaces. I'd guess which one makes it into the indexes is determined by the order in which the <refname> values appear.

There are some other cases such as stream wrappers (eg. bzip2:// and zlib://)