Extract Secp384r1 fast reduction from the prototype

[yanesca]

Extract Secp384r1 fast reduction from the prototype (replace the pre-existing implementation):
https://github.com/hanno-becker/mbedtls/blob/ecp_prototype/library/ecp_curves.c#L10172-L10262

This function calculates the residue of the input mod secp384r1_p. The result will have the same bitlength as the modulus. (That is, it is not guaranteed that it will be canonical.) This function is intended for reducing the result of a multiplication and therefore the number of input limbs can't be more than twice the limbs in the modulus.

Add unit tests for ecp_mod_p384_raw().

Tip for testing: fix up the result to get a canonical value by using #6375.