Work out why 90% of sbws measurements fail

longclaw's most recent vote shows that 90% of measurement attempts fail: recent_measurement_failure_count is 299K, and recent_measurement_attempt_count is 327K.

We should work out why sbws is doing so many failing measurements.

bandwidth-file-headers timestamp=1559442886 version=1.4.0 destinations_countries=ZZ earliest_bandwidth=2019-05-28T02:35:11 file_created=2019-06-02T02:35:03 generator_started=2019-05-19T14:04:34 latest_bandwidth=2019-06-02T02:34:46 minimum_number_eligible_relays=3934 minimum_percent_eligible_relays=60 number_consensus_relays=6556 number_eligible_relays=6287 percent_eligible_relays=96 recent_consensus_count=120 recent_measurement_attempt_count=327183 recent_measurement_failure_count=299072 recent_measurements_excluded_error_count=876 recent_measurements_excluded_few_count=678 recent_measurements_excluded_near_count=237 recent_measurements_excluded_old_count=0 recent_priority_list_count=991 recent_priority_relay_count=327183 scanner_country=US software=sbws software_version=1.1.0 time_to_report_half_network=225229
bandwidth-file-digest sha256=UkxK9KS5KZ5hKDiLI3bqGoMvpMW9gBjKGoYbD2bdZVE