<!-- .github/pull_request_template.md -->

## 📌 Description

The codegen logic for pytorch and tvm should unify after #1641 , and
this PR cleans up the related codegen functions in tvm_bindings.

Other changes:
1. update tvm-ffi to 0.1.0b11 to incorporate
apache/tvm-ffi#67 and
apache/tvm-ffi#68
2. rename of source files: `_ops.cu` and `_pybind.cu` renamed to
`_binding.cu`
3. remove torch related header include/library linking in ninja files
(#1642 (comment))
4. remove the use of `use_torch_stream` in unittests, they are no longer
required after apache/tvm-ffi#68

## 🔍 Related Issues

#1641 

## 🚀 Pull Request Checklist

Thank you for contributing to FlashInfer! Before we review your pull
request, please make sure the following items are complete.

### ✅ Pre-commit Checks

- [ ] I have installed `pre-commit` by running `pip install pre-commit`
(or used your preferred method).
- [ ] I have installed the hooks with `pre-commit install`.
- [ ] I have run the hooks manually with `pre-commit run --all-files`
and fixed any reported issues.

> If you are unsure about how to set up `pre-commit`, see [the
pre-commit documentation](https://pre-commit.com/).

## 🧪 Tests

- [ ] Tests have been added or updated as needed.
- [ ] All tests are passing (`unittest`, etc.).

## Reviewer Notes

cc @MasterJH5574 please let us know what changes do we need to make to
help you bump to the latest version of flashinfer in MLC.

Adds a pure tree-walking interpreter for `tvm_ffi.pyast` expression trees. The
evaluator never calls `compile`/`eval`/`exec`, never synthesizes source text,
and never round-trips through the stdlib `ast` module.

- Class-level dispatch table `_EXPR_EVALUATOR_DISPATCH: {pyast-node-type: handler}`
  drives `ExprEvaluator.eval`.
- `OperatorDispatch` keyed by `(op_kind, operand_type, operand_index)`; `lookup`
  walks operand MRO left-to-right; `invoke` falls back to `_NATIVE_HANDLERS`
  (a `{OperationKind: operator-module-callable}` table).
- `And`/`Or`/`IfThenElse`/`Parens`/`ChainedCompare` bypass user dispatch entirely
  for short-circuit correctness; unary ops go direct to native; binary ops route
  through `dispatch.invoke`. `ChainedCompare` is flattened to pairwise
  comparisons combined with `and` semantics.
- Scope is `Mapping[str, Any]`; internally wrapped in a `ChainMap` so
  `WalrusExpr` and per-lambda/per-comprehension bindings layer over but do not
  mutate the caller's mapping.
- `_build_generator` evaluates the outermost iterable eagerly (matching Python
  genexp semantics) and iterates remaining clauses lazily inside a generator.
- `Yield`/`YieldFrom`/`AwaitExpr` are recognized but raise `EvaluationError` on
  evaluation — valid tree shapes the evaluator refuses to execute.
- `eval_assign` rejects `Attr`/`Index` targets (evaluator only produces new
  local bindings) and supports at most one `StarredExpr` per sequence with
  matching trailing/middle capture.

Re-exported from `tvm_ffi` (purely additive, no renames or removals):

- `eval_expr`
- `eval_assign`
- `OperatorDispatch`
- `DEFAULT_DISPATCH`
- `EvaluationError`
- `UndefinedNameError`

None — runtime library additions only.

None. No existing public API is renamed, modified, or removed.

Design doc lives outside the repo; no Sphinx docs added in this commit.

- `tests/python/test_pyast_evaluator.py`: 112 tests covering every handler,
  dispatch MRO, short-circuits (`and`/`or`/chained compare), comprehension
  laziness (eager outermost iter, lazy body for generator expressions),
  starred/double-starred unpacking in calls/lists/dicts, walrus, f-strings
  (conversion flags + `format_spec`), `eval_assign` (including starred
  middle/end, Attr/Index rejection), and parity with native `eval` across
  representative sources. Includes a public-API re-export sanity check.
- `uv run pytest tests/python/test_pyast_evaluator.py` → 112 passed.
- `uv run pytest tests/python/test_pyast_evaluator.py tests/python/test_pyast.py
  tests/python/test_pyast_from_py.py` → 548 passed earlier in the session.
- `pre-commit run --files python/tvm_ffi/_pyast_evaluator.py
  tests/python/test_pyast_evaluator.py python/tvm_ffi/__init__.py` → ruff-check
  and ruff-format pass; ty has pre-existing `pytest` unresolved-import warnings
  affecting every test file in the repo; one `invalid-argument-type` for
  `ChainMap(dict, Mapping)` at 4 sites suppressed with `ty: ignore`; one
  `invalid-assignment` at the walrus-into-ChainMap line suppressed similarly.
  No new ty regressions introduced.

- Lambdas with defaults / `*args` / `**kwargs` (evaluator only supports plain
  positional `Id` params; the helper strips `:` and `=` suffixes but rejects
  `StarredExpr` params).
- Comprehension targets that are `Attr`/`Index`.
- Side-effectful `ChainMap` interaction from external callers.
- No fuzz testing.
- F-string debug specifier (`{x=}`) not explicitly tested.

This reverts commit 563cfd5.