fix: advance last_consolidated on consolidation failure to prevent re…

[TTdogCC]

…try storm

Root cause analysis of loop.py / memory.py:

When memory consolidation fails (LLM timeout or error), last_consolidated was not updated, causing the next trigger to retry an ever-growing batch. This leads to unbounded message accumulation and cascading LLM timeouts.

Specifically:

• last_consolidated is only updated on success — if the LLM call fails, it stays at the old position
• The trigger condition len(session.messages) > self.memory_window keeps firing every message once exceeded
• Result: 76 → 240 message accumulation observed in a single session, causing downstream LLM timeouts on the main agent loop

Fix: In the except block of _consolidate_memory, advance last_consolidated to the current position even on failure, matching the success-path logic. The if not archive_all guard ensures the /new command path is unaffected.

Note: concurrency guards (self._consolidating set + task lock) were already in place in loop.py.

Fixes #1174