docs: remove redundant Attributes: entries and drop TASK.md (#613)

Remove Attributes: sections where fields duplicate constructor params
verbatim (same name, same type, same description), keeping only entries
that add genuinely new information (computed/derived attributes):

- mellea/backends/backend.py: FormatterBackend
- mellea/core/base.py: CBlock, ImageBlock, ModelOutputThunk, ContextTurn
- cli/eval/runner.py: InputEvalResult, TestEvalResult (keep computed
  properties passed_count, total_count, pass_rate)

Also removes TASK.md left over from development.

Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>

Author: planetf1

TASK.md

@@ -33,12 +33,6 @@ class InputEvalResult:
         score (int): Numeric score assigned by the judge (``1`` for pass, ``0`` for fail).
         validation_reason (str): Justification text returned by the judge model.
 
-    Attributes:
-        input_text (str): The raw input text sent to the generation model.
-        model_output (str): The text response produced by the generation model.
-        validation_passed (bool): Whether the judge scored this response as passing.
-        score (int): Numeric score assigned by the judge (``1`` for pass, ``0`` for fail).
-        validation_reason (str): Justification text returned by the judge model.
     """
 
     def __init__(
@@ -81,10 +75,6 @@ class TestEvalResult:
             produced by running the generation and judge models.
 
     Attributes:
-        test_eval (TestBasedEval): The unit test specification containing
-            the test ID, name, instructions, inputs, and expected targets.
-        input_results (list[InputEvalResult]): Per-input evaluation outcomes
-            produced by running the generation and judge models.
         passed_count (int): Number of inputs that received a passing score.
         total_count (int): Total number of inputs evaluated.
         pass_rate (float): Fraction of inputs that passed (``passed_count / total_count``).

cli/eval/runner.py

@@ -34,10 +34,6 @@ class FormatterBackend(Backend, abc.ABC):
         model_options (dict | None): Default model options; if ``None``, an empty dict
             is used.
 
-    Attributes:
-        model_id (str | ModelIdentifier): The model identifier for this backend.
-        model_options (dict): The default model options for generation requests.
-        formatter (ChatFormatter): The formatter used to render components into prompts.
     """
 
     def __init__(

mellea/backends/backend.py

@@ -27,7 +27,7 @@
 from PIL import Image as PILImage
 
 
-class CBlock:
+canclass CBlock:
     """A `CBlock` is a block of content that can serve as input to or output from an LLM.
 
     Args:
@@ -36,9 +36,6 @@ class CBlock:
             completion object). Defaults to an empty dict.
         cache (bool): If ``True``, the inference engine may store the KV cache for this block. Experimental.
 
-    Attributes:
-        cache (bool): Whether the inference engine may store the KV cache for this block.
-        value (str | None): The underlying string content of the block.
     """
 
     def __init__(
@@ -89,8 +86,6 @@ class ImageBlock(CBlock):
         value (str): A valid base64-encoded PNG string (with or without a data URI prefix).
         meta (dict[str, Any] | None): Optional metadata to associate with this image block.
 
-    Attributes:
-        value (str): The base64-encoded PNG string representing the image.
     """
 
     def __init__(self, value: str, meta: dict[str, Any] | None = None):
@@ -272,10 +267,6 @@ class ModelOutputThunk(CBlock, Generic[S]):
         parsed_repr (S | None): An already-parsed representation to attach; set when re-wrapping existing output.
         tool_calls (dict[str, ModelToolCall] | None): Tool calls returned by the model alongside the text output.
 
-    Attributes:
-        parsed_repr (S | None): The parsed representation of the output, non-``None`` once computed.
-        tool_calls (dict[str, ModelToolCall] | None): Tool calls requested by the model, if any.
-        value (str | None): The raw string output from the model, or ``None`` if not yet computed.
     """
 
     def __init__(
@@ -555,11 +546,6 @@ class ContextTurn:
         output (ModelOutputThunk | None): The model's output thunk for this turn,
             or ``None`` for an input-only partial turn.
 
-    Attributes:
-        model_input (CBlock | Component | None): The input component or content block for this turn,
-            or ``None`` when the turn represents an output-only partial turn.
-        output (ModelOutputThunk | None): The model's output thunk for this turn,
-            or ``None`` when the turn represents an input-only partial turn.
     """
 
     model_input: CBlock | Component | None