fix(cli): surface memory backend start failures on stderr

[icn5381]

Summary

The four memory-backend start() call sites (agent REPL x2, gateway, TUI) caught failures with logger.exception only. Under the TUI loguru is redirected to a file, so an invalid memory identity (rejected by EverosBackend._validate_identity in start()) silently turned long-term memory off with nothing on screen. The agent REPL and gateway already surface the traceback via stderr; only the TUI was silent.

Add a warn_memory_start_failed helper that prints to stderr (mirroring the embedding-unavailable warning in everos/backend.py) and call it from every start() call site, so the degraded-memory condition is visible in every entry point.

Type

• Fix

Verification

• Relevant tests pass locally: uv run pytest tests/test_memory_warn.py -v -> 1 passed
• Relevant lint / type checks pass locally: uv run ruff format (5 files unchanged), uv run ruff check (all passed)
• User-facing docs or screenshots updated: N/A (internal warning, no doc surface)

Risk

• Security impact considered: adds a stderr notice only; no new code path, no shell, reads only the exception type name
• Backward compatibility considered: existing logger.exception calls are unchanged; the warning is additive
• Rollback path is clear: revert restores the four bare except blocks

Related Issues

Fixes #271