Fix a logic error in deterministic PRNG with buffer overrun issue

[concise]

In uECC_sign_deterministic, when uECC_WORD_SIZE is 8, the array declaration for T becomes uint64_t T[3]; and the following code

    #if (uECC_CURVE == uECC_secp160r1)
        T[uECC_WORDS] &= 0x01;
    #endif

becomes

        T[3] &= 0x01; // FIXME: access an array element outside its boundary?

I think that is not what we want (to keep only 161 pseudo-random bits so that 0 <= k <= 2^161-1) and the code in that case should be

        T[2] &= 0x00000001ffffffff; // keep the lower 33 bits

The entire code block may be changed into something like this:

#if (uECC_CURVE == uECC_secp160r1)
    #if (uECC_WORD_SIZE == 1)
        T[uECC_N_WORDS - 1] &= 0x01;
    #elif (uECC_WORD_SIZE == 4)
        T[uECC_N_WORDS - 1] &= 0x00000001;
    #elif (uECC_WORD_SIZE == 8)
        T[uECC_N_WORDS - 1] &= 0x00000001ffffffff;
    #endif
#endif

[concise]

Note that if #51 is accepted, this pull request can be discarded and closed.

[concise]

Oh, maybe I was doing nonsense. I didn't realized the usage for uECC_WORD_SIZE being 8 when compiling secp160r1 was forbidden for a long time:

#if (uECC_CURVE == uECC_secp160r1 ||
     uECC_CURVE == uECC_secp224r1) && (uECC_WORD_SIZE == 8)
    #undef uECC_WORD_SIZE
    #define uECC_WORD_SIZE 4
    #if (uECC_PLATFORM == uECC_x86_64)
        #undef uECC_PLATFORM
        #define uECC_PLATFORM uECC_x86
    #endif
#endif

Yet, since the macros for domain parameters of secp160r1 when uECC_WORD_SIZE is 8 are still there, I guess the fix might be useful in the future in general.

[kmackay]

I am planning to change micro-ecc to support runtime curve selection (see the 'runtime' branch). I think the implementation in that branch is correct for 8-byte words already.

You may want to base your RFC 6979 work on the runtime branch instead of master, since I will be replacing master with the runtime version as soon as it is ready (I need to add some more optimizations and AVR support).