tka: use ts_omit_tailnetlock in another spot, for ed25519consensus

I noticed this while modularizing clientupdate. With this in first,
moving clientupdate to be modular removes a bunch more stuff from
the minimal build + tsnet.

Updates #17115

Change-Id: I44bd055fca65808633fd3a848b0bbc09b00ad4fa
Signed-off-by: Brad Fitzpatrick <bradfitz@tailscale.com>

tka/key.go

@@ -8,7 +8,6 @@ import (
 	"errors"
 	"fmt"
 
-	"github.com/hdevalence/ed25519consensus"
 	"tailscale.com/types/tkatype"
 )
 
@@ -136,24 +135,3 @@ func (k Key) StaticValidate() error {
 	}
 	return nil
 }
-
-// Verify returns a nil error if the signature is valid over the
-// provided AUM BLAKE2s digest, using the given key.
-func signatureVerify(s *tkatype.Signature, aumDigest tkatype.AUMSigHash, key Key) error {
-	// NOTE(tom): Even if we can compute the public from the KeyID,
-	//            its possible for the KeyID to be attacker-controlled
-	//            so we should use the public contained in the state machine.
-	switch key.Kind {
-	case Key25519:
-		if len(key.Public) != ed25519.PublicKeySize {
-			return fmt.Errorf("ed25519 key has wrong length: %d", len(key.Public))
-		}
-		if ed25519consensus.Verify(ed25519.PublicKey(key.Public), aumDigest[:], s.Signature) {
-			return nil
-		}
-		return errors.New("invalid signature")
-
-	default:
-		return fmt.Errorf("unhandled key type: %v", key.Kind)
-	}
-}