feat: support creating BaguaTensor by passing torch tensor directly

[NOBLES5E]

No description provided.

[todo]

also check recv buf?

bagua-core/bagua-core-internal/src/communicators/mod.rs

Lines 390 to 395 in 13af9dd

	// TODO: also check recv buf?
	assert_eq!(
	send_tensor.num_elements_allocated() % self.nranks,
	0,
	"tensors must be aligned before using allscatter"
	);

---

This comment was generated by todo based on a TODO comment in 13af9dd in #19. cc @BaguaSys.

[todo]

separate crate for this

bagua-core/bagua-core-internal/src/torch_ffi.rs

Lines 1 to 6 in 13af9dd

	// TODO: separate crate for this
	#![allow(non_upper_case_globals)]
	#![allow(non_snake_case)]
	#![allow(dead_code)]
	pub const DeviceType_CPU: DeviceType = 0;
	pub const DeviceType_CUDA: DeviceType = 1;

---

This comment was generated by todo based on a TODO comment in 13af9dd in #19. cc @BaguaSys.

[todo]

sanity check

bagua-core/bagua-core-py/src/lib.rs

Lines 131 to 136 in 13af9dd

	// TODO: sanity check
	let dtype = torch_tensor
	.getattr("dtype")
	.expect("must pass valid torch tensor")
	.repr()?
	.to_string();

---

This comment was generated by todo based on a TODO comment in 13af9dd in #19. cc @BaguaSys.

[todo]

simplify args

bagua-core/bagua-core-internal/src/datatypes/mod.rs

Lines 731 to 734 in 9b549fc

	// TODO: simplify args
	pub fn get_communication_tensor(
	&self,
	stream_ptr: u64,

---

This comment was generated by todo based on a TODO comment in 9b549fc in #19. cc @BaguaSys.