fix: add nvidia-resiliency-ext to default dependencies (#2228)

Signed-off-by: Terry Kong <terryk@nvidia.com>

Author: terrykong

pyproject.toml

@@ -55,6 +55,7 @@ dependencies = [
   "cuda-bindings",                # for non-colocated refit
   "pybase64",                     # for sglang refit
   "nvidia-cudnn-cu12==9.19.0.56", # for transformer-engine no build isolation
+  "nvidia-resiliency-ext",        # for ft_launcher (fault-tolerant training launcher)
 ]
 
 [project.optional-dependencies]
@@ -160,6 +161,7 @@ megatron-core = { workspace = true }
 nemo-automodel = { path = "3rdparty/Automodel-workspace/Automodel", editable = true }
 megatron-bridge = { path = "3rdparty/Megatron-Bridge-workspace", editable = true }
 nemo_gym = { workspace = true }
+nvidia-resiliency-ext = { index = "pypi" }
 nemo_run = { git = "https://github.com/NVIDIA-NeMo/Run", rev = "414f0077c648fde2c71bb1186e97ccbf96d6844c" }
 # torch/torchaudio/torchvision/triton all come from the torch index in order to pick up aarch64 wheels
 torch = [