Merge pull request !28929 from ZengZitao/heter_compute_fix

Merge pull request !28784 from chenlei_autodiff/lite_callback

Merge pull request !28921 from jiaoy1224/aot

Merge pull request !28664 from jiaoy1224/aot

Merge pull request !28814 from ZengZitao/rewrite_output_nullptr

Merge pull request !28771 from TronZhang/recompute_fix_threshold

Merge pull request !28410 from chenlei_autodiff/lite_callback

Merge pull request !28256 from chenlei_autodiff/gk_lite

Merge pull request !27703 from DeshiChen/1214_litegraph

Merge pull request !28242 from jiaoy1224/aot

Merge pull request !27909 from 杨林枫/graph_kernel_opt

remove the "node_name" argument from class Node's constructor,
rename the node's "name_" to "debug_name_", and add the SetDebugName interface.

refactor the `OpRegistry`:
register the nodes by macro `OP_REGISTER`, instead of write them
in OpRegistry's constructor.

remove the five CompareOp's subclass, which did not override any virtual functions.
they can be binded to the `CompareOp` class.

Merge pull request !27588 from chenlei_autodiff/gk_lite

Merge pull request !28048 from r1chardf1d0/master

Merge pull request !27781 from chenlei_autodiff/lite_callback

the output address of internal output nop node is required by the next KernelGraph,
so we cannot replace it in graphkernel optimization.

Merge pull request !27456 from jiaoy1224/aot

Merge pull request !27454 from DeshiChen/1208_internal_output

Merge pull request !27394 from chenlei_autodiff/gk_lite

The internal output nodes of KernelGraph are stored in a `map`
for memory reuse between KernelGraphs.

However, the nodes may be changed in graphkernel optimization.
the passes of graphkernel are base on `FuncGraph`, instead of `KernelGraph`,
so we cannot use the `ReplaceInternalOutput` interface.

This commit, we add an attribute for these nodes and skip them in graphkernel's passes.

Merge pull request !27323 from DeshiChen/1206_elim_params

Merge pull request !27300 from hewei/core_api

1. Sequeue --> Sequence
2. Interger --> Integer