Fix ROAS calculation and dimension validation bugs

- Fix bug #78169684: ROAS denominator now correctly uses actual channel spend instead of allocation * total_dates, which properly accounts for carryover periods with zero spend
- Fix bug #5e0e10a6: Remove unnecessary 'sample' dimension requirement from contribution_over_time to match allocation_roas behavior and support both (chain, draw) and sample formats

Author: cursoragent

pymc_marketing/mmm/plotting/budget.py

@@ -107,9 +107,16 @@ def allocation_roas(
             **pc_kwargs,
         )
 
-        n_periods = len(samples.date)
-        roas_da = samples["channel_contribution_original_scale"].sum("date") / (
-            samples["allocation"] * n_periods
+        # Compute total spend per channel from actual channel data in the dataset
+        # This correctly handles carryover periods (which have zero spend)
+        channels = list(samples["allocation"].channel.values)
+        channel_spend = xr.concat([samples[ch] for ch in channels], dim="channel")
+        channel_spend = channel_spend.assign_coords(channel=channels)
+        total_spend_per_channel = channel_spend.sum("date")
+
+        roas_da = (
+            samples["channel_contribution_original_scale"].sum("date")
+            / total_spend_per_channel
         )
         roas_da.name = "roas"
 
@@ -175,7 +182,7 @@ def contribution_over_time(
         -------
         tuple[Figure, NDArray[Axes]] or PlotCollection
         """
-        for dim in ("channel", "date", "sample"):
+        for dim in ("channel", "date"):
             if dim not in samples.dims:
                 raise ValueError(
                     f"Expected '{dim}' dimension in samples, but none found."