Dataset test coverage to 80%

[osterbit]

(moving qri-io/qri#227 to this repo)

• After looking at codecov, it looks like dataset.vals has the most uncovered lines, and inside that, values.go is the biggest culprit.
• values.go has several functions that are expected to trigger a panic on failure (crash the program instead of just returning an error), which requires a slightly different testing template than usual
  • this example from jsonschema's schema_test provides a good template to follow for these types of functions (checks the value held by recover inside a deferred function following a panic):

func TestMust(t *testing.T) {
	defer func() {
		if r := recover(); r != nil {
			if err, ok := r.(error); ok {
				if err.Error() != "unexpected end of JSON input" {
					t.Errorf("expected panic error to equal: %s", "unexpected end of JSON input")
				}
			} else {
				t.Errorf("must paniced with a non-error")
			}
		} else {
			t.Errorf("expected invalid call to Must to panic")
		}
	}()

	// Valid call to Must shouldn't panic
	rs := Must(`{}`)
	if rs == nil {
		t.Errorf("expected parse of empty schema to return *RootSchema, got nil")
		return
	}

	// This should panic, checked in defer above
	Must(``)
}