Reduce and correct virtual buffer verbosity when moving by character and word

[jteh]

The current level of verbosity in virtual buffers when moving by character and word is insane. Hearing in/out of paragraph, text frame, section, etc. is pretty useless to users. Also, table and list information isn't reported correctly in this case. When entering a field, moving by character and word should read the same way as it does when moving by line, rather than special casing; e.g. "table of y rows and x columns" rather than "in table" and "separator" rather than "in separator". When leaving a field, it should only read "out of x" for very specific fields, such as links, form fields and maybe graphics.

[jteh]

If we are going to read when entering a heading, graphic, etc. when moving by char/word, maybe "out of" should report for those fields as well. Rationale: you can tell when you're leaving a heading when moving by line because a heading can only span one line, and if it spans multiple lines, NVDA announces heading again. When moving by char/word, you have no such guarantee, so you can't tell when you've left the heading.
The other option is to not read entry for headings, graphics, etc. when moving by char/word, but I'm not sure I like this.

[mdcurran]

(In #319) Hopefully we can fix in ticket 490

[jteh]

Bzr branch:  http://bzr.nvaccess.org/nvda/ticket490

[jteh]

(In #49) I'm rewriting most of this code in #490.

[jteh]

I've largely rewritten and simplified speech.getControlFieldSpeech(). Mick, I'd appreciate some comments on the approach; I think it is quite a bit nicer. Most of the old functionality should be reimplemented, plus the improvements in verbosity. I still need to fix an issue with reading on focus/quick nav; fix coming soon.

[jteh]

Merged in r3493.