Pair of tt-metal uplift improvements (new test, remove dep on libfmt.…

…so) (#1315)

* Remove libfmt.so.11 as dependency in ttrt build (setup.py)
 - Metal no longer builds its dependencies and it is said we don't need it.
* Update Silicon/TTNN/simple_reductions.mlir with case to repro tenstorrent/tt-metal/issues/15167
 - Originally found by tt-forge-fe training test during tt-metal uplift

runtime/tools/python/setup.py

@@ -66,7 +66,7 @@
     linklibs += ["TTRuntimeTTMetal", "tt_metal"]
 
 if enable_ttnn or enable_ttmetal:
-    runlibs += ["libdevice.so", "libnng.so.1", "libuv.so.1", "libfmt.so.11"]
+    runlibs += ["libdevice.so", "libnng.so.1", "libuv.so.1"]
     linklibs += ["TTRuntimeSysDesc", "TTRuntimeDebug", "TTRuntimeWorkarounds"]
 
 if enable_perf:

test/ttmlir/Silicon/TTNN/simple_reductions.mlir

@@ -17,6 +17,12 @@ func.func @sum_last_2_dims(%arg0: tensor<1x32x512x64xbf16>) -> tensor<1x32xbf16>
   return %1 : tensor<1x32xbf16>
 }
 
+func.func @sum_first_dim(%arg0: tensor<64x10xf32>) -> tensor<1x10xf32> {
+  %0 = tensor.empty() : tensor<1x10xf32>
+  %1 = "ttir.sum"(%arg0, %0) <{dim_arg = [-2 : i32], keep_dim = true, operand_constraints = [#any_device, #any_device]}> : (tensor<64x10xf32>, tensor<1x10xf32>) -> tensor<1x10xf32>
+  return %1: tensor<1x10xf32>
+}
+
 func.func @mean(%arg0: tensor<1x1x512x64xbf16>) -> tensor<1x1x512xbf16> {
   %0 = tensor.empty() : tensor<1x1x512xbf16>
   // CHECK: %[[C:.*]] = "ttnn.mean"[[C:.*]]