Add support for gpt-4.1 and gpt-4.5 model variants to encoding maps

[TaylorN15]

This PR adds support for newly released OpenAI model variants:

• gpt-4.1, gpt-4.1-mini, gpt-4.1-nano
• gpt-4.5, gpt-4.5-mini, gpt-4.5-nano

These models are not currently recognized by tiktoken.encoding_for_model() and raise an exception. All use the o200k_base tokenizer, so this patch ensures they are mapped appropriately.

Changes:

• Updated MODEL_TO_ENCODING for bare-name support (gpt-4.1, gpt-4.5)
• Updated MODEL_PREFIX_TO_ENCODING for all suffix variants (gpt-4.1-*, gpt-4.5-*)
• Extended test_encoding_for_model in test_misc.py to validate correct encoding resolution

Notes

• These changes follow the existing mapping pattern for gpt-4o and similar models.
• One unrelated test (test_hyp_roundtrip[cl100k_base]) fails due to tokenizer restrictions on special tokens (e.g., <|endofprompt|>), and is not impacted by this PR.

[TaylorN15]

I didn't notice #396, but this PR handles 4.1 and 4.5 and related tests.

[TaylorN15]

@hauntsaninja FYI