fix(stdlib): vectorized map properly handles shadowed columns

[jsternberg]

The vectorized version of map did not properly handle shadowed columns
the way the row based one worked. This copies over some of that code so
that it follows the correct logic. It also copies over the same logic
that is used to determine the column type when type inference does not
properly tell us the type.

This also fixes a bug in building objects that caused a bug in the
compiler. When a with is evaluated, the attributes from the original
record are copied over first and then overwritten values are assigned.
This overwriting did not properly release the values that were being
overwritten resulting in a memory leak.

Fixes #4812.

Done checklist

• docs/SPEC.md updated
• Test cases written

[Marwes]

I fixed the test I mentioned on slack so that it reproduces the error so if you rebase on #4817 and add this test we have a test to cover that. You seem to have fixed another issue as well though so if you can add a test for that also this looks good 👍

testcase drop_fn_duplicate_column_issue_4812 {
    got = array.from(rows: [{ x: 1, y: "a" }])
        |> map(fn: (r) => ({ r with x: r.x }))
        |> drop(columns: ["y"])

    want = array.from(rows: [{ x: 1 }])

    testing.diff(got, want)
}

[jsternberg]

Added the test case and also modified some of the code because the new code failed a test.