Feature requrest: enforce memory contraints

[dderjoel]

re bitcoin-core/secp256k1#1329 (comment)

• r and a may point to the same object, but neither can be equal to b. (...)
  */
  static void secp256k1_fe_mul(secp256k1_fe *r, const secp256k1_fe *a, const secp256k1_fe *b);

The idea is that we could have three enforcement levels:

• do not care at all (current default). This assumes, that all memory is distinct. I.e read a[0] after writing r[0] is fine.
• level 2: assume that r and a can be equal, but must be aligned. then, writing to r[0] and subsequently reading from a[1] is valid. Reading a[0] after writing r[0] would be invalid.
• level 3: do not read any memory after writing any. (except for stack). I.e. there cannot be any reads after the first write, expect if the read is from the stack.

[dderjoel]

implemented after 0c2922e

• Level 1: ./CryptOpt --memroyConstraints none
• Level 2: ./CryptOpt --memroyConstraints out1-arg1
• Level 3: ./CryptOpt --memroyConstraints all