Performance improvements

• Stops writing each existing record to an individual mrc file on disk. Instead caches each record's file and start/stop byte offsets in that file. This allows quick retrieval without all of the disk writing and still without holding everything in memory. This seems to speed up processing by about ~30%.
• Hashes marc.to_s for each record early on. When comparing incoming/existing records, first compares the hash values. Only when the hash values differ to we need to do more elaborate comparisons (e.g. omitting fields from comparison per specs, normalization etc.) to determine whether the record has changed. This might speed up processing of a set by another ~30%.
• Allows for conditionally adding MARC fields with parameters, using MarcEdit#add_conditional_field_with_parameters. See lines in config.yaml for example spec. The example adds a 590 to records when a 773 is found containing certain values. Further, the spec maps the 773 value into a parameter to include in the 590.
• Fixes gh-7. When dupe records exist in the existing set, they are no longer reported as being in the incoming set.