Check that user dependencies exist before running builders.

IMHO it's more intuitive if env.depends("my-target", "invalid-dep")
causes the build to fail when "invalid-dep" does not exist, rather
than having no effect (i.e. the build still succeeds with no warning).

This will make it more obvious, earlier, if you made a mistake in
an env.depends() call.  Otherwise the mistake could sit latent for
a while until you happen to notice that a build used a out-of-date
copy of some file rather than rebuilding it.

I've only added the check for user dependencies, not for direct
sources.  Listing a source file that doesn't exist will probably
cause the builder to fail anyway, so I don't think the preliminary
check is needed in that case.

spec/build_tests_spec.rb

@@ -417,14 +417,14 @@ EOF
     expect(result.stderr).to eq ""
     expect(lines(result.stdout)).to eq(["LD simple.exe"])
 
-    File.unlink("program.ld")
-    result = run_test(rsconsfile: "user_dependencies.rb")
-    expect(result.stderr).to eq ""
-    expect(lines(result.stdout)).to eq ["LD simple.exe"]
-
     result = run_test(rsconsfile: "user_dependencies.rb")
     expect(result.stderr).to eq ""
     expect(result.stdout).to eq ""
+
+    File.unlink("program.ld")
+    result = run_test(rsconsfile: "user_dependencies.rb")
+    expect(result.stderr).to match /User dependency program\.ld of target simple\.exe is invalid/
+    expect(result.status).to_not eq 0
   end
 
   unless ENV["omit_gdc_tests"]