Add base65536 and hexagram encoding injection probes

[assinscreedFC]

What

Adds two new encoding-injection probes to probes.encoding, addressing two items from the "extra encodings" wishlist in #152:

• InjectBase65536 — encodes payloads with Base65536 (2 bytes per character).
• InjectHexagram — encodes payloads as I Ching hexagrams (Base64 with each character mapped to a hexagram so the bits are visible).

Both follow the existing EncodingMixin pattern (primary_detector = encoding.DecodeMatch).

Dependencies

No new runtime dependency is introduced. Both encoders are clean-room ports of the MIT-licensed reference implementations, placed in resources/encodings.py with attribution comments and SPDX-FileCopyrightText lines:

• base65536_encode/decode — ported from qntm/base65536 (MIT). Validated byte-for-byte against the reference package.
• hexagram_encode/decode — ported from ferno/hexagram-encode (MIT). Matches the reference README vector (bytes([0x00,0x55,0xFF]) -> ䷁䷣䷄䷀).

Non-duplication

No open PR addresses these encoders (checked base65536, hexagram, and #152 references in open PRs). This PR covers only two wishlist items; #152 stays open for the rest.

Tests

New tests/resources/test_encodings.py adds roundtrip, known-vector, and invalid-input tests for both codecs. The new probes are also covered by the existing parametrized tests/probes/test_probes_encoding.py suite.

pytest tests/resources/test_encodings.py tests/probes/test_probes_encoding.py
  -> 112 passed, 1 skipped
pytest tests/test_docs.py
  -> 663 passed
black --check garak/probes/encoding.py garak/resources/encodings.py tests/resources/test_encodings.py
  -> clean

Codec ports were additionally verified with the base65536 reference package uninstalled, confirming no dependency.

AI assistance

This change was AI-assisted (Claude). The human submitter has reviewed every changed line and run the tests above.