Performance: accessing a column from a row should not allocate an array

[justinmeiners]

In the column access function it creates an array and then filters it (creating another). If I decode 1000 records with 10 columns each, that's going to allocate 10,000 arrays!
As a user, I would expect this to be a "do nothing" operation.

You can implement the same algorithm much more directly:

1. Find the first index where the keys match. If it's not there, this is the failed to find case.
2. Starting from the first index, find the next index where the keys match. If it exists, this is the ambiguous case.
   Otherwise we can now return the column at the first index.

[jberkel]

Fixed in #1115