[cli] support --strict mode

[spencatro-pub]

Currently when any kind of hook processing failure occurs, any attribution that was in the payload that failed is permanently lost. #33 attempts to mitigate this by at least recording traces of hook blobs (if parsing the blob somehow fails, the blob is still readable on disk and can be inspected to discover what went wrong). It would also be useful to optionally surface these failures back to the caller / client / harness, so that the user knows to check for missing attribution data, if they are concerned about that sort of thing and configure the tool as such.