Handle quantification analysis of bi-implication (`<=>') goals correctly.

Estimated hours taken: 10

Handle quantification analysis of bi-implication (`<=>') goals correctly.
Previously we used to expand bi-implications before doing quantification
analysis, which stuffed up the results of quantification analysis for
those goals.  We need to do quantification analysis first, and only
then can we expand bi-implications.  In addition, elimination of double
negation needs to come after expansion of bi-implication, so I moved
that from make_hlds.m to purity.m.

compiler/hlds_goal.m:
	Add a new alternative to the HLDS goal type for bi-implications.
        Also add a new predicate negate_goal, for use by purity.m.

compiler/make_hlds.m:
	Don't expand bi-implication here, instead just use the new
	bi_implication/2 HLDS goal type.
	Don't eliminated double negation here.

compiler/quantification.m:
	Handle quantification for bi-implications.
	Expand bi-implications.

compiler/purity.m:
	Eliminate double negation.

compiler/hlds_out.m:
	Add code to print out bi-implication goals.

compiler/*.m:
	Trivial changes to handle the new bi_implication/2
	alternative in the HLDS goal type.

compiler/notes/compiler_design.html:
	Document the above changes.

tests/hard_coded/Mmakefile:
tests/hard_coded/quantifier2.m:
tests/hard_coded/quantifier2.exp:
	A regression test for the above change.

compiler/lambda.m

@@ -255,6 +255,9 @@ lambda__process_goal_2(call(A,B,C,D,E,F), GoalInfo,
 lambda__process_goal_2(pragma_c_code(A,B,C,D,E,F,G), GoalInfo,
 			pragma_c_code(A,B,C,D,E,F,G) - GoalInfo) -->
 	[].
+lambda__process_goal_2(bi_implication(_, _), _, _) -->
+	% these should have been expanded out by now
+	{ error("lambda__process_goal_2: unexpected bi_implication") }.
 
 :- pred lambda__process_goal_list(list(hlds_goal), list(hlds_goal),
 					lambda_info, lambda_info).