fix: NULL checks, switch fall-through, thread safety, buffer bounds

[somethingwithproof]

Summary

Fixes all 10 HIGH-severity findings from static analysis / code review.

Findings Addressed

HIGH-1: Missing NULL check after calloc for poller_items (poller.c)
Added die() guard after calloc at the poller_items allocation.

HIGH-2: Missing NULL check after calloc for snmp_oids (poller.c)
Added die() guard after calloc at the snmp_oids allocation.

HIGH-3: Missing NULL check after malloc for error_string, buf_size, buf_errors (poller.c)
Added combined NULL check for all three malloc calls before dereferencing.

HIGH-4: Switch fall-through in get_date_format (util.c)
Added break; after each case body. Every case was falling through to the next, so only the default format was ever used.

HIGH-5: db_get_connection NULL return not checked (poller.c)
Added NULL checks with die() for both local and remote database connections.

HIGH-6: db_escape output buffer overflow risk (sql.c)
Input is now always capped to (trim_limit / 2) - 1 before passing to mysql_real_escape_string, regardless of whether the escaped size exceeds max_size.

HIGH-7: db_column_exists column name unescaped (sql.c)
Added character validation loop: only alphanumeric and underscore characters are accepted. Invalid column names log an error and return FALSE.

HIGH-8: Thread-unsafe static seq in ping_icmp (ping.c)
Replaced mutex-protected seq++ with __sync_fetch_and_add(&seq, 1) (GCC atomic built-in). Variable declared volatile for visibility.

HIGH-9: Unbounded sprintf in poller_push_data_to_main (util.c)
Replaced all sprintf calls with snprintf bounded by remaining space in the HUGE_BUFSIZE buffer.

HIGH-10: snprintf uses wrong buffer size constant (poller.c)
Changed snprintf(poll_result, BUFSIZE, ...) to snprintf(poll_result, RESULTS_BUFFER, ...) at both locations to match the actual malloc(RESULTS_BUFFER) allocation from snmp_get.

Verification

• Build: make clean && make -> 0 errors, 0 warnings
• No new dependencies or API changes

Test plan

• Verify build on Linux (CI)
• Run polling cycle with SNMP devices to confirm no regressions
• Confirm date format logging uses correct format per setting

[somethingwithproof]

Consolidated into mega PR #522 for independent mergeability.