analyzer: use the memory-optimized version of CHA

Passing nil as the "initial" argument of vta.CallGraph uses the new
internal version of the cha algorithm which stores its output graph
implicitly.  This saves time and memory for large callgraphs.

analyzer/util.go

@@ -16,7 +16,6 @@ import (
 
 	cpb "github.com/google/capslock/proto"
 	"golang.org/x/tools/go/callgraph"
-	"golang.org/x/tools/go/callgraph/cha"
 	"golang.org/x/tools/go/callgraph/vta"
 	"golang.org/x/tools/go/packages"
 	"golang.org/x/tools/go/ssa"
@@ -205,9 +204,8 @@ func buildGraph(pkgs []*packages.Package, populateSyntax bool) (*callgraph.Graph
 	}
 	ssaProg, _ := ssautil.AllPackages(pkgs, ssaBuilderMode)
 	ssaProg.Build()
-	graph := cha.CallGraph(ssaProg)
 	allFunctions := ssautil.AllFunctions(ssaProg)
-	graph = vta.CallGraph(allFunctions, graph)
+	graph := vta.CallGraph(allFunctions, nil)
 	return graph, ssaProg, allFunctions
 }