[SWDEV-539119] [release/2.8] Add fast_tanh support (#2484)

Perf improvement for triton tanh

(cherry picked from commit 4febbd8)

Author: jataylo

.ci/docker/ci_commit_pins/triton.txt

@@ -1 +1 @@
-5e5685356b9fc7b5ad9cdf4e510a1994a5b8601a
+6581064e5be2ac4ab0c198c1bb0e533a1900bbe3

torch/_inductor/codegen/triton.py

@@ -1306,7 +1306,7 @@ def tan(x):
     @staticmethod
     @maybe_upcast_float32()
     def tanh(x):
-        return f"libdevice.tanh({x})"
+        return f"libdevice.fast_tanhf({x})"
 
     @staticmethod
     @maybe_upcast_float32()