Optimize curve25519 field element by removing array

[paragonie-security]

By leveraging PHP's optimized object key access for what is effectively always a fixed-length array, we can remove some computational overhead and memory usage in by the PHP interpreter.

Since these memory access patterns are not secret (they were array indices from 0 to 9), the risk of side-channel leakage by this change is at least as low as what was previously implemented.

Unit Test Suite Runtime Ccomparison

CI Run	PHP 8.1	PHP 8.2	PHP 8.3	PHP 8.4	PHP 8.5
This Pull Request	Time: 00:28.808, Memory: 22.00 MB	Time: 00:28.185, Memory: 20.00 MB	Time: 00:28.469, Memory: 20.00 MB	Time: 00:29.143, Memory: 20.00 MB	Time: 00:28.431, Memory: 18.00 MB
CI for v2.3.1	Time: 00:36.502, Memory: 22.00 MB	Time: 00:33.545, Memory: 20.00 MB	Time: 00:33.295, Memory: 20.00 MB	Time: 00:32.855, Memory: 20.00 MB	Time: 00:34.210, Memory: 18.00 MB
CI for v2.3.0	Time: 00:36.373, Memory: 22.00 MB	Time: 00:32.035, Memory: 20.00 MB	Time: 00:33.618, Memory: 20.00 MB	Time: 00:32.713, Memory: 20.00 MB	Time: 00:33.804, Memory: 18.00 MB

I need to backport this change to v1.x and see if we get similar performance wins from the same strategy there (without sacrificing PHP 5.2 support) before I merge this pull request.

[paragonie-security]

The performance results from #199 are also telling:

PHP Version	#199	v1.22.0	v1.21.1
5.3	1.92 minutes	2.11 minutes	2.12 minutes
5.4	1.54 minutes	1.8 minutes	1.71 minutes
5.5	1.33 minutes	1.64 minutes	1.64 minutes
5.6	1.24 minutes	1.54 minutes	1.53 minutes
7.0	38.43 seconds	46.38 seconds	44.93 seconds
7.1	31.86 seconds	36.24 seconds	35.78 seconds
7.2	41.27 seconds	45.9 seconds	45.77 seconds
7.3	00:41.152	00:45.740	00:46.385
7.4	00:39.549	00:43.711	00:45.834
8.0	00:34.710	00:41.857	00:39.286
8.1	00:31.464	00:38.130	00:38.373
8.2	00:31.056	00:34.393	00:33.866
8.3	00:33.630	00:37.106	00:36.239
8.4	00:32.698	00:34.855	00:35.049
8.5	00:32.682	00:34.838	00:37.018