[Bug] broadcast无法指定tmp_ub的地址，在复用ub时会比较麻烦

[liuwenda4]

Bug Description

A clear and concise description of what the bug is.
broadcast无法指定tmp_ub的地址，在复用ub时会比较麻烦,在前端只有dst和src，还有axis，没有ub地址的参数，然而ascendc有这个选项

Environment

A2，A3
Hardware:

• NPU Device: [e.g. Ascend A2, A3]
• Other hardware info: [e.g. CPU, memory]

Software:

• OS: [e.g. Ubuntu 20.04, EulerOS]
• CANN version: [e.g. 8.3.RC1]
• torch-npu version: [e.g. 2.6.0.RC1]
• Python version: [e.g. 3.10]
• TileLang-Ascend version: [e.g. commit hash or tag]
• Installation method: [wheel package / build from source]

Code to Reproduce

# Minimal reproducible code
import tilelang
import torch

# Your code here

Error Message

Paste the full error message here

Expected Behavior

A clear and concise description of what you expected to happen.
broadcast可以指定tmp_ub

Actual Behavior

A clear and concise description of what actually happened.
broadcast没有这个tmp_ub的选项

Additional Context

Programming Mode:

• [x ] Developer mode (automatic sync/buffer management)
• [x ] Expert mode (manual T.Scope/T.barrier_all)
• [x ] Mixed mode

Backend:

• [x ] Ascend C & PTO backend
• AscendNPU IR backend

Pass Configs (if applicable):

pass_configs = {
    tilelang.PassConfigKey.TL_ASCEND_AUTO_SYNC: ...,
    tilelang.PassConfigKey.TL_ASCEND_MEMORY_PLANNING: ...,
    # ...
}

Operator Type:
[e.g. GEMM, Flash Attention, Softmax, Convolution, etc.]

Additional Information:

• Have you checked the Programming Guide?
• Have you searched existing issues?
• Any other context or screenshots?

[ArmandAlbert]

There`s a same ISSUE #1164 about the application of tmp buffer. Currently, a feasible solution is to modify pass and support manual allocation of the tmp buffer.

[ArmandAlbert]

PR #1256 has added this feature. Please check.