[KernelGen][Kunlunxin] Add digamma operator

[NineAnnAnn]

Summary

Adds a Kunlunxin (P800) backend specialization for digamma. Computes the logarithmic derivative of the gamma function element-wise on the input tensor. Adds a dedicated Triton kernel for this backend and registers it in the backend ops module.

Testing

• Validated against reference on device (Kunlunxin (P800))
• Accuracy: PASS (36 cases) over 3 dtypes x 6 shapes

Performance

Test command: pytest benchmark/test_digamma.py --level core (Kunlunxin (P800))

digamma

dtype	Size	Torch Latency (ms)	Gems Latency (ms)	Speedup	TFLOPS
float16	[64, 64]	0.184460	0.008181	22.547	0.001
float16	[4096, 4096]	59.264362	1.440451	41.143	0.012
float16	[64, 512, 512]	58.909332	1.440531	40.894	0.012
float32	[64, 64]	0.156772	0.006792	23.082	0.001
float32	[4096, 4096]	80.619667	1.347136	59.845	0.012
float32	[64, 512, 512]	80.892136	1.347404	60.036	0.012
bfloat16	[64, 64]	0.433782	0.007068	61.373	0.001
bfloat16	[4096, 4096]	63.654781	1.565451	40.662	0.011
bfloat16	[64, 512, 512]	62.773350	1.564299	40.129	0.011

Operator	Geometric Mean Speedup
digamma	40.825

Multi-backend Testing

Backend	Accuracy Test	Speedup (mean)	Notes
Kunlunxin (P800)	PASS (36 cases)	40.825	Primary

Files Changed

• src/flag_gems/runtime/backend/_kunlunxin/ops/digamma.py: New Kunlunxin (P800) kernel implementation
• src/flag_gems/runtime/backend/_kunlunxin/ops/__init__.py: Register import and __all__ entry

[Dongxu-H]

/test | digamma:kunlunxin

[github-actions]

❌ On-demand test failed

Operator: digamma
Runner: kunlunxin
Backend: kunlunxin

The test failed to complete. Check the workflow run for details.

[Dongxu-H]

code-style CI failed
all 中的 digamma 放在 zeros_like 之后不符合字母顺序，应放在 digamma_ 之前
logger 应使用 Kunlunxin backend 约定写法而非主文件夹写法