CPP: Cache Expr.getType().

[geoffw0]

@jbj as discussed in the meeting. I tested this with a series of five queries from a clean cache (largeparameter.ql, castarraypointerarithmetic.ql, badlyboundedwrite.ql, wrongtypeformatarguments.ql, slicing.ql, all of which are on LGTM), with the first run discarded.

Chakracore
base:
	largeparameter.ql 38s
	castarraypointerarithmetic.ql 233s
	badlyboundedwrite.ql 31s
	wrongtypeformatarguments.ql 29s
	slicing.ql 8s
	TOTAL: 339s
with cached `Expr.getType()`:
	largeparameter.ql 35s
	castarraypointerarithmetic.ql 206s
	badlyboundedwrite.ql 13s
	wrongtypeformatarguments.ql 23s
	slicing.ql 3s
	TOTAL: 280s (83%)

Qt
base:
	largeparameter.ql 79s
	castarraypointerarithmetic.ql 470s
	badlyboundedwrite.ql 31s
	wrongtypeformatarguments.ql 59s
	slicing.ql 9s
	TOTAL: 648s
with cached `Expr.getType()`:
	largeparameter.qls 79s
	castarraypointerarithmetic.ql 457s
	badlyboundedwrite.ql 27s
	wrongtypeformatarguments.ql 57s
	slicing.ql 6s
	TOTAL: 626s (97%)

Note that there are more queries on LGTM that use this predicate.

The worrying part is that Expr.getType() should be rather trivial:

pragma[nomagic] Type getType() { expr_types(underlyingElement(this),unresolveElement(result),_) }

so it's a bit surprising it showed up so high in our profiling, even being so heavily used.

Other predicates from the spreadsheet that might be in a similar situation to this one: Element::getLocation, Expr::toString, Call::getTarget, Element::getFile, Location::hasLocationInfo, Access::getTarget, Class::getCanonicalMember.

[jbj]

Expr.getType is non-trivial because it's overridden in FunctionCall. I'd still like to investigate if we can speed it up, but caching it seems like a good choice in any case.