Log recovery and salvage should handle removal or truncation of log files

This is related to the work in WT-3276.  If a log file that will be used in recovery (i.e. past the checkpoint) is removed or truncated, then we do not get expected recovery. In the case of truncation, recovery will appear to stop at the corrupted file with an error, and then a subsequent start with recovery will succeed, but no records past the checkpoint are visible. This is the current behavior without any changes for WT-3276. Recovery should in fact fail, and continue to fail until salvage runs.  Salvage should leave the log in a state where records up to the point of corruption are visible.

The work for WT-3276 contains a new test_txn19.py that has an NOTE: comment to highlight the trouble area.