Can be -1 for NUL-terminated string.  This is useful for passing parts
of a larger string to a function without having to copy or modify the
string first.

Affected functions:

	hb_tag_t hb_tag_from_string()
	hb_direction_from_string()
	hb_language_from_string()
	hb_script_from_string()

As reported by Khaled on the list:

"After the introduction of canonical reordering of combining marks
(commit 34c22f81), I'm no longer able to do mark/mark substitution or
positioning for mark sequences that involve shadda as a first mark (or
most interesting sequences at least).

"After some digging, it turned out that shadda have a ccc=33 while most
Arabic marks that combine with it have a lower ccc value, which results
in the shadda being reordered after the other mark which,
unsurprisingly, breaks my contextual substitution and mkmk anchors."

See:

http://unicode.org/faq/normalization.html#8
http://unicode.org/faq/normalization.html#9

Oops!

I don't see how this function can be useful.

To be modified, a lot.

Fixes build with MSVC.

need to figure out the naming.  The generated code doesn't have the
right name.

For two reasons:

1. User can always call hb_buffer_pre_allocate() themselves, and

2. Now we do a pre_alloc in add_utfX anyway, so the total number of
reallocs is limited to a small number (~3) anyway.  This just makes the
API cleaner.

Using graphite2's comparerenderer suggests that this makes hb-ft 15
times faster.  No caching layer needed anymore.

We do a conservative estimate of the number of characters, but still,
this limits the number of buffer reallocs to a small constant.

According to Peter Constable this is indeed what Uniscribe has been
doing for years.

Mozilla Bug 667166 - wrong shape of letter when it comes at the end of
word in the arabic version of Firefox 5.0

If the specified shapers fail, hb-shape will fail immediately