fix(anthropic): consecutive assistant message merge drops content on mixed types

[teknium1]

Summary

In agent/anthropic_adapter.py, when merging consecutive assistant messages (required by Anthropic's API which rejects consecutive same-role messages), the mixed-types case (one message has string content, the other has list content) simply replaced the earlier message with the later one:

# Before (line 967):
fixed[-1] = m  # Earlier message content silently dropped

This caused data loss when the conversation had consecutive assistant messages with different content formats.

What changed

Applied the same normalize-and-merge pattern already used in the tool_use merge path (lines 952-956): convert both to [{"type": "text", "text": ...}] list format before concatenating.

# After:
if isinstance(prev_blocks, str):
    prev_blocks = [{"type": "text", "text": prev_blocks}]
if isinstance(curr_blocks, str):
    curr_blocks = [{"type": "text", "text": curr_blocks}]
fixed[-1]["content"] = prev_blocks + curr_blocks

Test plan

• Trigger consecutive assistant messages with mixed content types (e.g. tool call producing list content followed by text response)
• Verify both messages' content is preserved in the merged result