Separate content identity from file location in execution evidence #60

The xFit and xScan executors and device pipeline include absolute paths and file modification times in configuration identity. Identical inputs mounted at a different path therefore produce different configuration hashes, preventing comparisons across sites.

Follow up on Carmelo's reviews of #42 and #56: define content identity from input/checkpoint digests and computation options, while recording paths and timestamps separately as provenance. Preserve corruption and changed-input detection. Add cases showing that relocated identical inputs have equal content identity and changed bytes or computation options do not.

This needs an explicit evidence-schema compatibility decision and belongs in a separate change from #56's MPI planning fix.