based on #942

Implements EdDSA with the avx2/avx512 ed25519 implementations, resulting
in a 2.8x speedup. The speedup is also in part to a slightly faster
straus implementation than the one the zig stdlib uses, since it can be
variable time + different radix setup.

```
+------------------+-------+-------+-------+----------+
| naiveBatchVerify |  min  |  max  |  mean |  std_dev |
+==================+=======+=======+=======+==========+
| _                |  3719 |  5888 |  3863 |  186     |
+------------------+-------+-------+-------+----------+
+----------------+-------+-------+-------+----------+
| stdBatchVerify |  min  |  max  |  mean |  std_dev |
+================+=======+=======+=======+==========+
| _              |  1579 |  2510 |  1625 |  73      |
+----------------+-------+-------+-------+----------+
+----------------+-------+-------+-------+----------+
| sigBatchVerify |  min  |  max  |  mean |  std_dev |
+================+=======+=======+=======+==========+
| _              |  1315 |  2107 |  1349 |  45      |
+----------------+-------+-------+-------+----------+
```

Unfortunately it is not currently possible to fully batch transaction
verification. Agave loops over transaction signatures and
`verify_strict`s each one, in order to tighten down on the behaviour
regarding low-order points. While this makes some sense, in reality it
greatly reduces our ability to do sigverify fast, and doesn't really
have any benefit to the protocol. Wish Solana could use something like
ed25519-zebra instead.