fix(notebooks): surface exception type on empty-message failures

[nathantournant]

Problem

Notebook sync failures can surface an empty error message, leaving operators with no diagnostic signal:

Sync-CLI type=notebooks (took 32.4s) failed -- error summary (1 lines):
... - ERROR - [notebooks - <id>] -

The trailing dash + space + newline is literal — the error body is empty, not truncated.

Root cause: bare aiohttp.ServerTimeoutError() and asyncio.TimeoutError() have empty __str__. They propagate through custom_client.py (which doesn't translate them) into resources_handler.py's except Exception as e block, where the handler logs str(e) directly.

Fix

New helper format_exc_for_log(exc) in datadog_sync/utils/resource_utils.py:

• Timeout family (asyncio.TimeoutError, aiohttp.ServerTimeoutError) → timeout: <ClassName>[: <msg>]
• Non-empty str(exc) → returned verbatim (no regression for CustomClientHTTPError, SkipResource, ResourceConnectionError, etc.)
• Empty str(exc) → exception class name

Applied at four failure-log sites in resources_handler.py:

• _apply_resource_cb — the exact site that emits the empty-error log line.
• _import_resource — promotes detail from DEBUG-only to ERROR (was invisible at default verbosity).
• _force_missing_dep_import_cb (×2) — consistency.
• _cleanup_worker — consistency.

Tests

11 new tests in tests/unit/test_notebooks_error_formatting.py:

• 6 helper-level unit tests covering the three branches.
• 5 ResourcesHandler integration tests pinning the empty-error / timeout / non-empty-pass-through / HTTP-error-preserved contracts.

Verification

• pytest tests/unit/ — 629 passed, 0 failed.
• ruff check + black --check — clean.

Out of scope

• The ~30s timeout itself. This PR makes the failure observable; tuning the per-notebook timeout is a separate question. Now that the type is surfaced, an operator can decide whether the right answer is a longer timeout, batched requests, or per-notebook payload-size limits.

Follow-ups

• Investigate the timeout root cause once this PR is in and the type is visible in logs.
• Consider extending the same format_exc_for_log helper to other resource models if their error paths show the same shape.