fix(blocks): skip input validation in dry-run mode for blocks with sentinel credentials

Two fixes for dry-run execution of nested agents:

1. _base.py: Skip validate_data() when execution_context.dry_run is True.
   prepare_dry_run() sets credentials=None for OrchestratorBlock (platform
   key injected separately), but the block's own JSON schema validation
   rejected None as "required property". This caused any dry-run execution
   of graphs containing OrchestratorBlock to fail with BlockInputError.

2. agent.py: Check required inner-agent inputs against data["inputs"]
   instead of top-level data keys (previous commit 6f03ceeb88).

autogpt_platform/backend/backend/blocks/_base.py

@@ -698,13 +698,20 @@ class Block(ABC, Generic[BlockSchemaInputType, BlockSchemaOutputType]):
             if should_pause:
                 return
 
-        # Validate the input data (original or reviewer-modified) once
-        if error := self.input_schema.validate_data(input_data):
-            raise BlockInputError(
-                message=f"Unable to execute block with invalid input data: {error}",
-                block_name=self.name,
-                block_id=self.id,
-            )
+        # Validate the input data (original or reviewer-modified) once.
+        # Skip validation in dry-run mode — sentinel credential values (None)
+        # would fail JSON schema required checks, and simulate_block /
+        # prepare_dry_run have already ensured the input is usable.
+        is_dry_run = getattr(
+            kwargs.get("execution_context"), "dry_run", False
+        )
+        if not is_dry_run:
+            if error := self.input_schema.validate_data(input_data):
+                raise BlockInputError(
+                    message=f"Unable to execute block with invalid input data: {error}",
+                    block_name=self.name,
+                    block_id=self.id,
+                )
 
         # Use the validated input data
         async for output_name, output_data in self.run(