fix: raise ValueError instead of exit(1) on missing API key; return error string from searxSearch on network failure

[octo-patch]

Problem

Two related error-handling bugs crash the backend or break the tool contract:

1. Provider.get_api_key() calls exit(1) when the API key is missing (sources/llm_provider.py:63). This terminates the entire backend process immediately — including when running inside Docker — instead of surfacing a clear error. Users then see the WebUI as offline with no useful diagnostics.

2. searxSearch.execute() raises an Exception on RequestException (sources/tools/searxSearch.py:107). All other execute() implementations return an error string; raising here breaks the tool API contract assumed by execute_modules() in agent.py, crashes the agent mid-task, and also causes test_execute_request_exception to fail (the test correctly expects a return value).

Solution

• Replace exit(1) in get_api_key() with raise ValueError(...) so callers receive a proper Python exception with a descriptive message rather than a silent process crash.
• Replace the raise Exception(...) in searxSearch.execute() with a return of an error string (prefixed with "Error during search:" so execution_failure_check() detects it correctly).

Testing

• python -m pytest tests/test_searx_search.py — all 9 tests pass, including test_execute_request_exception which now correctly receives an error string instead of an unhandled exception.
• python -m pytest tests/ — only the pre-existing test_save_and_load_memory failure remains (caused by a termcolor<2.4.0 installation in the test environment, unrelated to this PR).