This commit modifies the pre-computed table used to optimize the secp256k1
scalar multiplication to a string instead of a byte slice. This change
makes the compile more efficient since the Go compiler internally
represents bytes slices inefficiently.
This reduces the memory needed to compile btcec to 3MB versus the previous
40MB before this change.
In addition, it modifies the code which loads the pre-computed table to
deserialize directly into the table instead of into locals that are then
copied.
Fixes#297.