Update xla to use mlir rather than backend-specific-translations

[pseudo-rnd-thoughts]

Description

Fixes #313

Motivation and Context

EnvPool XLA doesn't work with Jax 0.4.29+

Types of changes

What types of changes does your code introduce? Put an x in all the boxes that apply:

• Bug fix (non-breaking change which fixes an issue)
• New feature (non-breaking change which adds core functionality)
• New environment (non-breaking change which adds 3rd-party environment)
• Breaking change (fix or feature that would cause existing functionality to change)
• Documentation (update in the documentation)
• Example (update in the folder of example)

[pseudo-rnd-thoughts]

@Trinkle23897 The linting error is unrelated to the PR

[pseudo-rnd-thoughts]

Testing this throws errors however there appears to be limited documentation on what the suggested change should be

[JagerHoHo]

After implementing the suggested changes, I encountered an error when running the XLA example with JAX 0.4.34(latest). The error reads:

TypeError: CustomCallWithLayout(): incompatible function arguments. The following argument types are supported:
    1. CustomCallWithLayout(builder: jaxlib.xla_extension.XlaBuilder, call_target_name: bytes, operands: Span[jaxlib.xla_extension.XlaOp], shape_with_layout: jaxlib.xla_extension.Shape, operand_shapes_with_layout: Span[jaxlib.xla_extension.Shape], opaque: bytes = b'', has_side_effect: bool = False, schedule: jaxlib.xla_extension.ops.CustomCallSchedule = CustomCallSchedule.SCHEDULE_NONE, api_version: jaxlib.xla_extension.ops.CustomCallApiVersion = CustomCallApiVersion.API_VERSION_ORIGINAL) -> jaxlib.xla_extension.XlaOp

Invoked with types: jax._src.interpreters.mlir.LoweringRuleContext, bytes, kwargs = { operands: tuple, operand_shapes_with_layout: tuple, shape_with_layout: jaxlib.xla_extension.Shape, opaque: bytes, has_side_effect: bool }

I installed envpool using pip install envpool and manually applied the changes as instructed. It seems there may be an issue with how the CustomCallWithLayout is invoked in the current context with the latest JAX.

envpool/envpool/python/xla_template.py

Lines 74 to 88 in f411fc2

	def translation(c: Any, *args: Any, platform: str = "cpu") -> Any:
	output_shape_with_layout = _shape_with_layout(out_specs)
	if len(out_specs) == 1:
	output_shape = output_shape_with_layout[0]
	else:
	output_shape = xla_client.Shape.tuple_shape(output_shape_with_layout)
	return xla_client.ops.CustomCallWithLayout(
	c,
	f"{type(obj).__name__}_{id(obj)}_{name}_{platform}".encode(),
	operands=args,
	operand_shapes_with_layout=_shape_with_layout(in_specs),
	shape_with_layout=output_shape,
	opaque=handle,
	has_side_effect=True,
	)

Would appreciate any guidance on resolving this.