Report non-retryable PayloadValidationError as BAD_REQUEST

[bergundy]

What changed

A payload codec or payload converter can now signal that a Nexus operation's input is invalid by raising a non-retryable ApplicationError of type PayloadValidationError while decoding the input. Such an error is translated into a BAD_REQUEST nexusrpc.HandlerError with the message Invalid operation input, retaining the original error as its __cause__.

Previously such an error became an INTERNAL handler error on the codec path, which callers retry — so a caller sending invalid input was retried until timeout instead of failing fast. On the converter path the type was already BAD_REQUEST, but the message was indistinguishable from a genuine decode failure; it now gets its own message.

Unchanged

• ApplicationError of any other type → INTERNAL on the codec path (as before)
• A retryable PayloadValidationError → INTERNAL (non-retryable is required)
• The external-storage retrieve path and the generic decode-failure messages are untouched
• The serialize/output path is untouched — BAD_REQUEST would be wrong for a result-encoding failure

Tests

tests/nexus/test_workflow_caller_errors.py covers both the codec and converter stages: positive unit cases asserting BAD_REQUEST, not retryable, the wrapper message and the preserved cause; negative cases for a different error type and for a retryable PayloadValidationError; plus an end-to-end test asserting the caller sees a non-retryable BAD_REQUEST with an ApplicationError cause.

Cross-SDK

Part of a coordinated change; equivalent PRs exist for Go, Java, TypeScript and .NET. The wrapper message wording is aligned across SDKs, adapted to each SDK's message style.