ed25519: scalarmult fix so dst can be src

[ph4r05]

ge25519_scalarmult minor changes:

• Result r can be equal to the input p1 (same point pointer).
• Returns fully valid extended Edwards point (not partial as before).
  • Makes further operations easier because many operations expect full point (with valid T coordinate), not the partial.
  • ge25519_scalarmult_base_niels already returns full point so it would make it more consistent
  • Scalar multiplications perform large number of curve25519_mul, typically (64*const), one more before returning from the function is IMO small overhead.
  • if ge25519_scalarmult returns partial point one cannot easily make it full point because after ge25519_scalarmult returns, the temporary point is not not accessible. To make it full point much more expensive inversion would be needed.

[ph4r05]

Similar to
#172

[onvej-sl]

ACK

[prusnak]

Thx!