Enable building of the HTML manual by default unless not building from
Git and with an existing local copy (which is included in distribution
tarballs).

This makes sure we can install the HTML manual without having it
checked in VCS, yet not require rst2html for tarball builds.

As it is generated by the build system it doesn't have to be checked
in, and having it in became a whole lot more annoying since it started
being generated automatically on `make all` instead of specific (and
weird) `make doc`, as it means whenever the documentation changes the
HTML is re-generated on any make run.  This is particularly problematic
if using a different Docutils version than the one used to generate the
checked-in version as it would create local noise that should not be
committed, yet keep being annoying.

This situation probably leads to most people disabling (or at least,
not enabling) the documentation generation on normal builds, itself
leading to more hassle updating of the documentation.

Adding new keyword hold, used e.g. on SQLAnywhere to open a cursor 'with hold'

From a patch by TJF.  Part of [feature-requests:#691].

Make reflow paragraph leave cursor at end of reflowed text, which is more likely to be useful than where it was left at the start of the last line.

This makes the API more obvious on what argument is expected, and avoid
confusion between Geany and Scintilla find flags.

Passed-in flags was always 0, so the argument is not useful.
Also, this function expected Scintilla search flags rather than Geany
ones, making the API confusing for no good reason.

search_find_prev() wasn't properly updated to work on GeanyFindFlags
leading to incorrect flags handling.

Patch from TJF.  Part of [feature-requests:#691]

See http://www.freebasic.net/wiki/wikka.php?wakka=CatPgFunctIndex

Part of [feature-requests:#691]

Don't use the $LINGUAS at configure time to set ALL_LINGUAS, and let
the translations Makefile properly honor the $LINGUAS itself, which it
already does better than we do, cleanly ignoring translations it
doesn't know.

If we do set ALL_LINGUAS=$LINGUAS, it will result in a build failure if
we do not have a translation for some of the language(s) specified in
$LINGUAS, and would make it impossible to build other languages without
re-configuring.  So, just drop that part and let the build-time support
kick in.

Closes #507.

Patch partly from Yosef Or Boczko, thanks.  Closes #959.

Closes #1062.

`entries_modified` global is only used by on_name_entry_changed() and
on_entries_changed(), both of which are only ever called from
project_new() scope -- which already initializes this variable.

Having add_input_widgets() didn't really make the code any simpler,
rather obfuscating it a little.

Use gtk_dialog_run() to run the key input dialog, which is modal anyway.
This avoids having to pass the label and the iter around for the dialog
response callback to have them, as they now only are used directly in the
function setting them in the first place.

This however doesn't get rid of any of the global data itself, it only pack it
in a struct and passes pointer to this struct around instead of accessing the
global whenever possible.