Reduce moco notebook training iterations in CI via FAST_CI_MODE (#1542)

## Summary

Fixes
[BIO-403](https://linear.app/nvidia/issue/BIO-403/investigate-flaky-tests-and-timeout-errors-in-bionemo-moco-notebook):
Investigate flaky tests and timeout errors in bionemo-moco notebook.

The `discrete_data_interpolant_tutorial.ipynb` notebook has 3 training
loops that hard-code `range(50000)` iterations. When executed in CI via
`nbval`, these loops cause timeouts.

## Changes

1. Added a configuration cell that detects the `FAST_CI_MODE`
environment variable (already set by `unit-tests-framework.yml` when
running notebook tests)
2. Sets `NUM_TRAINING_STEPS = 500` when in CI mode, `50000` otherwise
3. Replaced all 3 `range(50000)` occurrences with
`range(NUM_TRAINING_STEPS)` in the DFM, D3PM, and MDLM training loops

This follows the existing pattern used in
`bionemo-recipes/recipes/evo2_megatron/examples/` notebooks.

## Testing

- The notebook behavior is unchanged when run outside CI (`FAST_CI_MODE`
not set)
- When `FAST_CI_MODE=true` (as in framework CI), iterations drop from
50,000 to 500, preventing timeouts

Signed-off-by: svc-bionemo <267129667+svc-bionemo@users.noreply.github.com>
Co-authored-by: svc-bionemo <267129667+svc-bionemo@users.noreply.github.com>

Author: svc-bionemo

sub-packages/bionemo-moco/examples/discrete_data_interpolant_tutorial.ipynb

@@ -22,6 +22,19 @@
     "torch.cuda.manual_seed(42)"
    ]
   },
+  {
+   "cell_type": "code",
+   "execution_count": null,
+   "metadata": {},
+   "outputs": [],
+   "source": [
+    "import os\n",
+    "\n",
+    "\n",
+    "FAST_CI_MODE: bool = os.environ.get(\"FAST_CI_MODE\", \"\").lower() in (\"1\", \"true\", \"yes\")\n",
+    "NUM_TRAINING_STEPS = 500 if FAST_CI_MODE else 50000"
+   ]
+  },
   {
    "cell_type": "markdown",
    "metadata": {},
@@ -144,7 +157,7 @@
    "source": [
     "model = model.to(DEVICE)\n",
     "losses = []\n",
-    "for _ in tqdm(range(50000)):\n",
+    "for _ in tqdm(range(NUM_TRAINING_STEPS)):\n",
     "    num_ones = torch.randint(0, D + 1, (B,))\n",
     "    x1 = (torch.arange(D)[None, :] < num_ones[:, None]).long().to(DEVICE)\n",
     "    # x1 e.g. [1, 1, 1, 0, 0, 0, 0, 0, 0, 0] or [1, 1, 1, 1, 1, 1, 1, 1, 1, 0]\n",
@@ -659,7 +672,7 @@
     "# NBVAL_SKIP\n",
     "model = model.to(DEVICE)\n",
     "losses = []\n",
-    "for _ in tqdm(range(50000)):\n",
+    "for _ in tqdm(range(NUM_TRAINING_STEPS)):\n",
     "    num_ones = torch.randint(0, D + 1, (B,))\n",
     "    x1 = (torch.arange(D)[None, :] < num_ones[:, None]).long().to(DEVICE)\n",
     "    # x1 e.g. [1, 1, 1, 0, 0, 0, 0, 0, 0, 0] or [1, 1, 1, 1, 1, 1, 1, 1, 1, 0]\n",
@@ -892,7 +905,7 @@
     "\n",
     "model = model.to(DEVICE)\n",
     "losses = []\n",
-    "for _ in tqdm(range(50000)):\n",
+    "for _ in tqdm(range(NUM_TRAINING_STEPS)):\n",
     "    num_ones = torch.randint(0, D + 1, (B,))\n",
     "    x1 = (torch.arange(D)[None, :] < num_ones[:, None]).long().to(DEVICE)\n",
     "    # x1 e.g. [1, 1, 1, 0, 0, 0, 0, 0, 0, 0] or [1, 1, 1, 1, 1, 1, 1, 1, 1, 0]\n",