Polish: Intellisense is a little too aggressive

[MaulingMonkey]

Edit: some of this was just from bugs: #293

This is a good problem to have ;). I'd like to gather some cases where intellisense should perhaps not trigger, or trigger with less information, partially by comparing against e.g. C#'s intellisense. I'm hopeful many of these can be solved with simple lexer checks. My partial list of differences I've discovered so far follows:

Differences

Rust	C#	Notes
let f	var f	Intellisense shouldn't try to autocomplete other identifiers when we're trying to define a new identifiers, although perhaps other keywords like mut would still be sane?
pub fn f	public static void f	Ditto
func(	func(	Intellisense shouldn't try to "help" us until we start typing a new identifier (signature help, on the other hand, is appropriate)
let a = (	var a = (	Ditto, minus the signature help.
foo.	foo.	Intellisense shouldn't display keywords when only identifiers will do

Equivalent

Rust	C#	Notes
let◄◄◄	var◄◄◄	Both seem to continue to display intellisense after typing an identifier occured, to my suprise.

Work list

• Don't autocomplete new identifiers
• Don't autocomplete newly opened argument lists (showing signature help instead is OK)
• Don't autocomplete newly opened parenthetical groups (nor trigger new signature help)
• Intellisense docs should display "summary", but not arguments, exceptions, etc. Rustdoc equivalent might be the first summary before any # ... headings.
• Do autocomplete after scoping operators (., ::)
• Signature help docs should also display exceptions, and the current argument. Rustdoc equivalent might be the addition of parsed # Arguments sections, and # Panics or # Failure sections.
• Generic types and methods should display <>s in the intellisense list, even if no <>s are autocompleted.

[MaulingMonkey]

Some work done to fix this in MaulingMonkey@7b9f314#diff-d34d4bfb25c66e3dbd8f55cc2c8e38d5R90

I still want to look at other cases where intellisense triggers to compare them to the C# implementation...

[MaulingMonkey]

My changes introduced regressions: . and :: structural tokens generally shouldn't inhibit intellisense, even though ( and < should.