rustdoc: Merge source code pages HTML elements together

[GuillaumeGomez]

We realized that the HTML generated for the source code pages could be improved quite a lot. This PR is a first pass toward this goal. Some explanations: it merges similar classes elements (even when there are white characters in between).

There is an exception to this: if this is an ident, I also merged it with "unclassified" elements. This part is up to debate and can be very easily removed as the check is performed in one place (in the can_merge function).

EDIT: The ident is now only kept in the code for the span it contains but it is not rendered into the HTML.

So now some numbers:

For these ones, on each page, I run this JS: document.getElementsByTagName('*').length. The goal is to count the number of DOM elements. I took some pages that seemed big, but don't hesitate to check some others.

file name	before this PR	with this PR	diff	without ident	diff
std/lib.rs.html (source link on std crate page)	3455	2776	20.7%	2387	31%
alloc/vec/mod.rs.html (source on Vec type page)	11012	8084	26.6%	6682	39.4%
alloc/string.rs.html (source on String type page)	10800	8214	24%	6712	37.9%
std/sync/mutex.rs.html (source on Mutex type page)	2953	2403	18.7%	2139	27.6%

You can test it here.

cc @jsha
r? @notriddle