ECC changes

[NIIBE Yutaka]

Hello,

When I visited Cologne last year, I addressed that we could
improve ECC implementation of libgcrypt.

This summer, I created a ticket for one thing: field specific
methods.

    https://dev.gnupg.org/T3358

And I did some work for 25519.  (We also can do for other curves.)

    https://dev.gnupg.org/source/libgcrypt/history/gniibe-T3358/

While it's faster than current implementation, please note that this is
still much slower than the original implementation of Ed25519/X25519.
The original implementation doesn't use MPI, but uses its own field
element representation.

Introducing different field element representations should be done in
future.  I think that some restructuring is needed for libgcrypt to do
that.
--