fix: correct bounds check in XCDA array access

[o7si]

In the constructor of llm_tokenizer_ugm, xcda_array is an array with xcda_array_size elements:

llama.cpp/src/llama-vocab.cpp

Lines 776 to 777 in 017ecee

	xcda_array = (const uint32_t *) &precompiled_charsmap[charsmap_offset];
	xcda_array_size = xcda_blob_size / sizeof(uint32_t);

The valid index range is [0, xcda_array_size).

When xcda_array_view is constructed using these parameters:

llama.cpp/src/llama-vocab.cpp

Line 1048 in 017ecee

struct xcda_array_view xcda_view(tokenizer.xcda_array, tokenizer.xcda_array_size);

The accessible index range remains [0, xcda_array_size).

Therefore, in the xcda_array_view::get_node function, it should use >= to check for bounds violation instead of >.

[CISC]

Nice catch, thank you, merging when CIs are done.