fix: preserve compatibility with old-style A2A3 TPipe local slot arg

[ndleslx]

Summary

• make include/pto/npu/a2a3/TPush.hpp accept both old-style and new-style TPipe template argument layouts
• interpret the 5th template argument as IsNoSplit only for 0/1, otherwise treat it as LocalSlotNum
• keep current behavior for new call sites while restoring compatibility with current ptoas v0.24 output

Problem

Current generated code from ptoas v0.24 may emit:

TPipe<0, Direction::DIR_C2V, 4096, 8, 8>

but the current A2A3 header declares the 5th template parameter as bool IsNoSplit, which makes that form fail to compile.

Fix

Use a compatibility parameter in the 5th position and derive:

• IsNoSplit from 0/1
• the effective local slot count from values >1

This preserves support for both:

• TPipe<..., SlotNum, LocalSlotNum>
• TPipe<..., SlotNum, IsNoSplit, LocalSlotNum>

Validation

• confirmed hw-native-sys/ptoas latest release is v0.24
• confirmed current hw-native-sys/pto-isa:main still had the incompatible A2A3 signature
• verified the patch matches the downstream compatibility workaround that unblocked ptoas v0.24 generated A2A3 kernels

Closes #67

[gemini-code-assist]

Code Review

This pull request refactors the TPipe template parameters in TPush.hpp to introduce CompatLocalSlotNumOrIsNoSplit, replacing the IsNoSplit boolean. This update allows for more flexible local slot number configurations while maintaining backward compatibility. Feedback was provided to simplify the boolean logic for the IsNoSplit constant to improve code clarity.

[gemini-code-assist]

The logic for IsNoSplit can be simplified. Since CompatLocalSlotNumOrIsNoSplit is a uint32_t, the expression CompatLocalSlotNumOrIsNoSplit <= 1 && static_cast<bool>(CompatLocalSlotNumOrIsNoSplit) is logically equivalent to CompatLocalSlotNumOrIsNoSplit == 1. This is more concise and improves readability while maintaining the same behavior (evaluating to true only when the value is 1).

    static constexpr bool IsNoSplit = (CompatLocalSlotNumOrIsNoSplit == 1);