TE Integration fails with TE main

[kshitij12345]

Running the tests with latest TE main, pytest thunder/tests/test_transformer_engine_executor.py results in

      t26 = Tensor.to(x, torch.bfloat16, copy=True)  # t26: "cuda:0 bf16[32, 32]"
      t28 = Tensor.to(w, torch.bfloat16, copy=True)  # t28: "cuda:0 bf16[32, 32]"
>     (t6, (t12, t13, t14, t15, t16, t17, t18, t19), ctx_te_2311) = te_linear_22(t26, t28, None)
E     ValueError: not enough values to unpack (expected 8, got 6)

thunder.computation_37:17: ValueError

NOTE: This can't be reproduced on pjnl-latest as it seems to be pinned to TE v2.3

[IvanYashchuk]

The problem is either in _get_num_saved_tensors or get_recipe_from_options_or_default_recipe:

lightning-thunder/thunder/executors/transformer_engineex.py

Line 337 in d0f6474

for _ in range(_get_num_saved_tensors(get_recipe_from_options_or_default_recipe()))

[kshitij12345]

Fixed in #2146