Add new regression and ensemble experiment configurations for V2-M and OrthoBridge

- Introduced multiple regression experiment configurations targeting vf_md, including:
  - cd_solo_reg_set.json: CD tower only regression setup.
  - img_solo_reg_set.json: Image tower only regression setup.
  - reg_head_epoch_sweep.json: Baseline regression sweeps at different epochs (50, 75, 100).
  - reg_head_set.json: Various regression setups including baseline and OrthoBridge configurations.
  - single_eye_reg.json: Single-eye regression setup for worst-eye aggregation analysis.

- Added ensemble configurations for OrthoBridge with different inner bridges:
  - ortho_alts_ensemble.json: Ensemble tests with ConcatBridge, PairwiseAdditiveBridge, and GatedAdditiveBridge.
  - ortho_alts_tritower.json: Tritower tests with the same inner bridges.

- Created V2-M specific configurations:
  - baseline_reg_nt50.json: Regression baseline with V2-M backbone.
  - geom_vec_gt.json and geom_vec_unet.json: Geometry vector injection experiments with V2-M.
  - single_l1_bridges.json: Single-eye ensemble experiments with various bridge types.
  - tritower_geom_gt.json: Tritower setup with GT contour-rasterized masks.

- Promoted existing experiments to higher repetitions for robustness.

v4/scripts/experiments/convnext/baseline_convnextv2_tiny.json

@@ -0,0 +1,15 @@
+[
+  {
+    "_note": "ConvNeXt-V2-Tiny vs the resnet50/refugelike img backbone. Otherwise identical to experiments/tri_v1/baseline_ensemble — same ensemble_fused base, binary classification, hb eval.",
+    "run_name": "experiments/convnext/baseline_convnextv2_tiny",
+    "reps": 10,
+    "tower_overrides": {
+      "img": {
+        "args": {
+          "backbone": "convnextv2_tiny",
+          "freeze_ratio": 0.0
+        }
+      }
+    }
+  }
+]

v4/scripts/experiments/convnext/diagnose_baseline_failure.json

@@ -0,0 +1,46 @@
+[
+  {
+    "_note": "Hypothesis 1 — ImageNet pretraining is OOD for fundus; freezing low-level filters helps. Keep tiny but freeze 60% of stages (stem + first 2 of 4 stages).",
+    "run_name": "experiments/convnext/baseline_convnextv2_tiny_freeze60",
+    "reps": 3,
+    "tower_overrides": {
+      "img": {
+        "args": {
+          "backbone": "convnextv2_tiny",
+          "freeze_ratio": 0.6
+        }
+      }
+    }
+  },
+
+  {
+    "_note": "Hypothesis 2 — model too big for 330 patients; try the smallest variant.",
+    "run_name": "experiments/convnext/baseline_convnextv2_atto",
+    "reps": 3,
+    "tower_overrides": {
+      "img": {
+        "args": {
+          "backbone": "convnextv2_atto",
+          "freeze_ratio": 0.0
+        }
+      }
+    }
+  },
+
+  {
+    "_note": "Hypothesis 3 — LR too high for the bigger model from ImageNet init; halve LR with the tiny backbone unfrozen.",
+    "run_name": "experiments/convnext/baseline_convnextv2_tiny_lr5e5",
+    "reps": 3,
+    "overrides": {
+      "training": { "lr": 5e-5 }
+    },
+    "tower_overrides": {
+      "img": {
+        "args": {
+          "backbone": "convnextv2_tiny",
+          "freeze_ratio": 0.0
+        }
+      }
+    }
+  }
+]