fix: migrate TVM API from tir to s_tir/tirx for apache-tvm 0.25

[WilliamOdinson]

Summary

apache-tvm 0.25 renamed/reorganized the TensorIR scheduling API. This PR migrates all assignment2 code to the new API so the assignment works with the current TVM release.

The schedule in schedule_example.ipynb was verified on Google Colab with a T4 GPU.

Note

My Colab install command is:

python -m pip install --pre -U -f https://mlc.ai/wheels mlc-llm-nightly-cu128 mlc-ai-nightly-cu128

API changes

Old (tvm <0.25)	New (apache-tvm 0.25)
from tvm import tir	from tvm import s_tir
from tvm.script import tir as T	from tvm.script import tirx as T
from tvm.tir.schedule import BlockRV	from tvm.s_tir.schedule import SBlockRV
from tvm import meta_schedule as ms	from tvm.s_tir import meta_schedule as ms
from tvm.meta_schedule.space_generator import ScheduleFn	from tvm.s_tir.meta_schedule.space_generator import ScheduleFn
tir.Schedule(...)	s_tir.Schedule(...)
@T.prim_func	@T.prim_func(s_tir=True)
T.alloc_buffer(...)	T.sblock_alloc_buffer(...)
T.block("name")	T.sblock("name")
sch.get_block("name")	sch.get_sblock("name")
T.thread_binding(N, "blockIdx.x")	T.thread_binding(N, thread="blockIdx.x")
tvm.nd.array(...)	tvm.runtime.tensor(...)
tvm.nd.empty(...)	tvm.runtime.empty(...)
tvm.cuda()	tvm.runtime.cuda() (or from tvm.runtime import cuda)
tvm.testing.assert_allclose(...)	np.testing.assert_allclose(...)
f.imported_modules[0].get_source()	f.imports[0].inspect_source()

Known limitation

reference.py is not included in this PR. As a student I don't have the correct solution to regenerate it against the new API; this file should be updated by course staff.