Performance optimisations on stitchIntoRecord

[far-blue]

This PR suggests some performance optimisations on stitchRecord, as per the discussion in #12. The general approach is:

• Exit the stitchIntoRecord method early when possible.
• Remove foreignRecord entries in stitchIntoRecord once they've been matched, where possible.

I've not analysed/improved many-2-many relationships in this PR but I suspect there is possible further improvement in the case of many-2-1 and many-2-many around keying foreign records on a hash of their match fields to prevent multiple scans through the foreignRecords array. However, this would require a larger re-working on the stitching methods.

[far-blue]

I'm nervous about these changes because they seem so obvious I've almost convinced myself I've missed something that means these improvements won't work! So please don't just take my word for it that these changes are good in all possible situations. They certainly appear correct in my application but that's not saying much.

All unit tests pass.

[pmjones]

@far-blue I have taken a quick look. The early-return portions look reasonable, but I am a little uneasy about the unsets. I will do some experimentation here and report back. Thank you very much for putting this together!

[pmjones]

@far-blue Yes, these all look reasonable. My unease with the unsets was unfounded. In 1:1 and 1:M relationships, the native primary key is specified on the foreign table, which means that no two natives can ever relate to the same foreign; thus, unsetting does not remove a possibly-related record. I expect to merge this today/tomorrow. Thanks again for your good work here!

[far-blue]

Thanks for accepting the PR :)