Research/solve truncating failure stack traces SDK-side so they don't get truncated server side

[cretz]

Describe the solution you'd like

Today if a failure is too large, the server truncates. But the server can't truncate a stack trace for an "encoded" failure (i.e. one that is in the encoded_attributes), so that entire set of payloads is dropped which means the entire message and stack trace is gone, not just partial. So we need to do this SDK side.

Need to truncate the stack trace to a reasonably small size by default, ideally cutting out the middle frames only replacing them with "...X frames omitted..." or something. And this should be overridable in the failure converter options. First task is to find out what that number is, which likely requires consulting with the server to see what their default truncation size is and stay a bit under that (since theirs may be whole-proto size and ours is just stack trace).

Per-SDK Tickets

• Go -
• Java -
• Core -
• TypeScript -
• Python -
• .NET -
• Ruby -
• PHP -
• Temporal CLI - N/A