tables.create / add_columns cannot set column constraints (MaxLength, MinValue/MaxValue, Format, RequiredLevel, DisplayName) #194

Summary

client.tables.create(...) and client.tables.add_columns(...) accept a column spec of {schema_name: type} where the value is only a type string (or an Enum subclass for optionsets). There is no way to set per-column constraints -- MaxLength, MinValue/MaxValue, Format/FormatName, Precision, RequiredLevel, or a custom DisplayName. Every constraint is hardcoded per type.

Evidence (main @ 5601657)

src/PowerPlatform/Dataverse/data/_odata_base.py::_attribute_payload:

• Rejects any non-str / non-Enum spec: raise ValueError("... expected str or Enum subclass").
• Hardcodes constraints per type: string -> MaxLength: 200; memo -> MaxLength: 4000; int -> MinValue: -2147483648, MaxValue: 2147483647; decimal/double -> fixed Min/Max + Precision; RequiredLevel: {"Value": "None"} on all branches; DisplayName derived from the schema-name suffix.

No public create_attribute / raw-metadata / generic-request method exists in the tables API (only create, add_columns, remove_columns, create_alternate_key), so there is no managed escape hatch either.

Impact

To create, for example, a Rating column limited to 1-5 or a Comment column with MaxLength=2000, callers must drop to the raw Web API EntityDefinitions({id})/Attributes endpoint and hand-build the payload -- including the @odata.type derived-type discriminator and using PUT (not PATCH) for updates. On hosts where the Dataverse CLI cannot run (ChatGPT web / Codex sandbox -- no .NET runtime), the SDK is the primary path, so this gap forces error-prone hand-rolled metadata calls for a very common ask ("create a table with a bounded rating and a sized text field").

Proposed change (backward-compatible)

Extend the column-spec value to also accept a dict, keeping str / Enum working exactly as today:

client.tables.create("cfb_CustomerFeedback", {
    "cfb_Name": "string",                                         # unchanged
    "cfb_Rating":  {"type": "int",  "min_value": 1, "max_value": 5},
    "cfb_Comment": {"type": "memo", "max_length": 2000, "display_name": "Comment"},
})

_attribute_payload builds the base payload for type, then applies the provided overrides:

• max_length -> MaxLength
• min_value / max_value -> MinValue / MaxValue
• precision -> Precision
• format -> Format (int/date) or FormatName.Value (string/memo)
• required -> RequiredLevel.Value (None / ApplicationRequired / Recommended)
• display_name -> DisplayName label

The @odata.type discriminator is already stamped correctly, so this needs no endpoint changes.

Touchpoints

• data/_odata_base.py::_attribute_payload (core -- shared by the sync + async paths).
• Docstrings / typing for operations/tables.py + aio/operations/async_tables.py (create, add_columns) and operations/batch.py.
• Unit tests mirroring the existing _attribute_payload type tests (add override cases).
• README + CHANGELOG per the repo maintenance rules.

Acceptance

• {"type": "memo", "max_length": 2000} produces a MemoAttributeMetadata payload with MaxLength: 2000.
• {"type": "int", "min_value": 1, "max_value": 5} produces an IntegerAttributeMetadata payload with those bounds.
• str and Enum specs behave exactly as before (no breaking change).
• Unit tests cover the overrides and the unchanged str/Enum paths.