* It is not necessary to subtract 'id' from the list of copied attributes because record and mem_record are equal, so therefore their id attributes are also equal (so there is no harm in copying it, and this reduces the complexity of the code)
 * It is not necessary to intersect the attribute names, since record and mem_record are equal, so they have the same id and class, so they have the same columns in the database. If record has non-column attributes then it seems reasonable to also copy them onto mem_record (though I am not sure what situation this would ever happen in)

CollectionAssociation#merge_target_lists should write to the underlying attributes when copying, rather than using the assignment method

Previously the runtime was reset implicitly when #cleanup_view_runtime was called at the end of most requests. However, this doesn't happen when the request redirects, or send_file is called.  Consequently, the ActiveRecord runtime recorded in the logs included the time taken for both the current request and the previous redirect.  Explicitly resetting at the start of each request ensures that this can't happen, no matter what occurs previously.

Remove pointless rescue (it doesn't happen anywhere in the tests, and I can't see a way that it would happen ever)

Don't use mass-assignment protection when setting foreign keys or association conditions on singular associations. Fixes #481 (again).

updated AR#create! to accept an options hash so the mass-assignment security role can be passed in, also updated the Changelog to mention the change to some of the AR method signatures.

- If doing a count on a relation that has an :include and a :join, it does a distinct even though it should not.
- When doing a count on a relation that has an :include, it always falls back to a old style left join when performing the count.  Looks like it was broken here:
https://github.com/rails/rails/commit/b9599502c9e738a5a1513e75d08f8d40ed408265

Add test to specify that attributes from an association's conditions should be assigned without mass-assignment protection when a record is built on the association.

Wrapping each Record.associations.build in a transaction is going to
make several unneeded queries. Reverting this commit also fixes #479.