[BACKEND] Add Metax McTLE support

[cjh9368]

METAX backend support for TLE Lite tile primitives in this patch

This patch adds METAX/MACA support for the main TLE Lite tile primitives:

• tle.extract_tile(x, index, tile_shape)

  • Extracts a tile of shape tile_shape from a register tensor at a tile-grid index.
  • Supported index forms:
    • Multi-dimensional static index: tuple/list of int/constexpr, linearized at compile time.
    • Scalar static index: int/constexpr, treated as the linearized tile index.
    • Scalar dynamic index: scalar i32/i64 tl.tensor.
    • Multi-dimensional dynamic index: tuple/list containing tl.tensor, row-major linearized in the frontend.
  • tile_shape must be compile-time constants, match the source rank, and divide the source shape evenly along each dimension.

• tle.insert_tile(x, tile, index)

  • Inserts a register tile into a source tensor at a tile-grid index.
  • The result type matches the source tensor type.
  • Index forms and constraints are the same as extract_tile.
  • The tile shape is derived from the tile operand and must divide the source shape evenly.

Lowering path

Key points for the METAX implementation:

• The Python frontend is moved under triton.experimental.tle for the METAX backend, matching the shared TLE frontend import path.
• The old third_party/metax/language/maca/mctle Python location is removed.
• Tile operations are emitted through the mctle pybind builder interfaces:
  • create_extract_tile
  • create_insert_tile
• The generated IR uses METAX-local mctle dialect ops:
  • mctle.extract_tile
  • mctle.insert_tile
• The METAX backend continues to load the mctle dialect and related lowering passes through triton._C.libtriton.mctle.

Scope

This PR only enables:

• tle.extract_tile
• tle.insert_tile

[CLAassistant]

All committers have signed the CLA.

[zhzhcookie]

Please format the code by pre-commit, and add tle tests in .github/workflows/metax3.6-build-and-test.yml.
Please update md5 in https://github.com/flagos-ai/FlagTree/blob/triton_v3.6.x/python/setup_tools/setup_helper.py#L698, and change url to metaxTritonPlugin-cpython3.12-x86_64_v0.6.1.tar.gz.

[cjh9368]

build