Bug: mine --mode convos silently skips claude.ai exports due to sender/role field mismatch and 10MB file size limit

Environment

• MemPalace version: 3.1
• Python: 3.13
• Platform: Home Assistant add-on (Linux/Docker)
• Export source: claude.ai privacy export (Settings → Export Data)

Bug 1: sender vs role field mismatch in normalize.py

_try_claude_ai_json checks item.get("role") to identify message authors, but the claude.ai privacy export format uses "sender" instead of "role".

Actual claude.ai export structure:

{
  "uuid": "...",
  "sender": "human",
  "content": [...],
  "text": "..."
}

_try_claude_ai_json looks for:

role = item.get("role", "")
if role in ("user", "human") and text:

Since role is always None, zero messages are extracted and zero drawers are filed — silently.

Suggested fix (normalize.py):

role = item.get("role", item.get("sender", ""))

Bug 2: 10MB file size limit silently skips large exports

convo_miner.py defines:

MAX_FILE_SIZE = 10 * 1024 * 1024  # 10 MB

A standard claude.ai export with ~130 conversations produces a conversations.json of ~38MB — nearly 4x the limit. The file is silently skipped in scan_convos with no warning or error message, making it very difficult to diagnose.

Workaround: Manually split conversations.json into batches of ~10 conversations before mining:

import json, os
with open('conversations.json') as f:
    data = json.load(f)
batch_size = 10
for i in range(0, len(data), batch_size):
    batch = data[i:i+batch_size]
    with open(f'convos_{i:03d}.json', 'w') as f:
        json.dump(batch, f)

Suggested fixes:

• Print a warning when a file is skipped due to size, e.g.:
  ⚠ Skipping large file (38MB > 10MB limit): conversations.json
• Raise the default limit — claude.ai exports are commonly 20–50MB
• Document the limit clearly in the README

Steps to Reproduce

1. Export data from claude.ai (Settings → Account → Export Data)
2. Place export files in a directory
3. Run mempalace mine /path/to/export/ --mode convos --wing claude-ai
4. Observe: 0 conversation drawers filed despite valid export files present

Expected Behaviour

All conversations in conversations.json are normalised and filed into the palace.

Actual Behaviour

• Bug 1: conversations.json is parsed but produces 0 messages due to sender/role mismatch — 0 drawers filed
• Bug 2: conversations.json (38MB) silently skipped before parsing due to 10MB size limit — no warning shown