secp256k1 version bump

[kristapsk]

Argument why use specific commit of secp256k1 was to have the same version as coincurve uses (https://github.com/JoinMarket-Org/joinmarket-clientserver/pull/536/files#r418937166). Since January coincurve uses f2d9aeae6d5a7c7fbbba8bbb38b1849b784beef7.

Related - fort-nix/nix-bitcoin#371.

[BitcoinWukong]

Verified:

-> ~ wget https://github.com/bitcoin-core/secp256k1/archive/f2d9aeae6d5a7c7fbbba8bbb38b1849b784beef7.tar.gz
-> ~ sha256sum f2d9aeae6d5a7c7fbbba8bbb38b1849b784beef7.tar.gz 
bdcb0373cc830271733841cf507ad2bb95d61be558f5334c8ad304d127ee4da9  f2d9aeae6d5a7c7fbbba8bbb38b1849b784beef7.tar.gz

[AdamISZ]

Notes on potential things to pay attention; first, these are commits since f2d9aeae6d5a7c7fbbba8bbb38b1849b784beef7 :
1f233b3fa05eb29a744487e0682d925055fb0d4c Remove num/gmp support - this is part of bitcoin-core/secp256k1#831 which removes libgmp and replaces the functions with new algorithms by Peter Dettman as I understand. This might be relevant if we have gmp flags set somewhere in our stack for compilation, and need to remove them.

99f47c20ec41279075d6b3ae64c9c1a84b40a6f8 Fixes problems with ARM builds(?) identified here. (Very unlikely to matter but just in case this impacts some users of our project).

Recent big changes to the Schnorr signing API, clearly not relevant.

From 0d9540b13ffcd7cd44cc361b8744b93d88aa76ba to f2d9aeae6d5a7c7fbbba8bbb38b1849b784beef7 :

Indeed, the only changes I see to the public API are in this PR which implemented Schnorr signing.

Also worthy of note is this PR which means ECDH module is no longer considered experimental.

Given this I doubt we will see any issues. It may be that the change in removing gmp creates a performance boost, but I don't think it'll be huge so no need to rush moving forward to a version with that, I think.