[Relax][NN] Use int64 for RoPE apply flag

[xthomaswang]

This patch aligns the dtype of the apply_rope flag used by
llama_rope_with_position_map with the host-side value passed through
Relax call_tir.

Previously the PrimFunc declared apply_rope as T.int32, while the
caller-side scalar value is represented as an int64 Relax PrimValue /
ShapeExpr value. This caused Relax well-formed analysis to reject the IR
with:

Argument N type mismatch: expected R.Prim("int32"), given R.Prim(value=1)

The mismatch can be reproduced through downstream nn.Module.export_tvm
paths such as MLC-LLM convert_weight / compile.

This change updates:

• llama_rope_with_position_map: apply_rope: T.int32 -> T.int64
• PagedKVCache: pass the split-rotary flag as int64_t

[gemini-code-assist]

Code Review

This pull request updates the data type of the apply_rope parameter from int32 to int64 in the Python frontend and changes the corresponding cast to int64_t in the C++ runtime. These changes ensure consistency in integer precision across the codebase. I have no feedback to provide.

[xthomaswang]

Downstream impact:

This fixes MLC-LLM export paths that instantiate the RoPE PrimFunc through
Relax nn.Module.export_tvm, including mlc_llm convert_weight and
mlc_llm compile. Without this patch, those paths may produce a
not-well-formed Relax IR warning or fail at the frontend well-formedness
assert, depending on whether the downstream tree demotes the assert.

[cchung100m]

LGTM, thanks to @xthomaswang 😄