fix(model): add exact mapping for gpt-4.5

[yen0304]

encoding_for_model("gpt-4.5") raises a KeyError today:

>>> import tiktoken
>>> tiktoken.encoding_for_model("gpt-4.5")
KeyError: 'Could not automatically map gpt-4.5 to a tokeniser. ...'

Root cause: MODEL_PREFIX_TO_ENCODING contains "gpt-4.5-" (note the trailing hyphen), which correctly handles versioned variants like gpt-4.5-preview. However, the bare model name "gpt-4.5" does not start with "gpt-4.5-" — the trailing hyphen is never matched — so it falls through to the KeyError path.

This is the same shape as the gpt-5.1 issue (#464 / PR #554). Every other dot-minor model (gpt-5, gpt-4.1, gpt-4o) has both an exact entry in MODEL_TO_ENCODING and a versioned prefix. gpt-4.5 was missing the exact entry.

Fix: add "gpt-4.5": "o200k_base" to MODEL_TO_ENCODING, consistent with all peer models.

Verified locally:

>>> tiktoken.encoding_for_model("gpt-4.5").name
'o200k_base'
>>> tiktoken.encoding_for_model("gpt-4.5-preview").name  # prefix still works
'o200k_base'

This contribution was made with AI assistance.