Fix expression column aggregate calculation with group_by after remove()

[texodus]

Fixes a bug which causes expression columns to show null or 0 results when a remove() had taken place before.

When a View with expressions is created from a Table that has had remove() applied to it, some aggregates (like sum) are calculated incorrectly, as in the example below. Interestingly, this issue only occurs when the remove() occurs before the view(). If all remove() calls occur after the view is created, a slightly different code path correctly calculates the expressions.

Ultimately this issue was caused by a few under-tested code paths, in which he wrong mapping was chosen deep in a giant block of data processing code., given a choice between the "original" and remove()-corrected indices when matching primary keys to column's row index. The repro (below), simple as it is, was instrumental in identifying and fixing it.

schema = {"a": str, "b": float, "c": str}
data = [
    {
        "a": "A",
        "b": 46412.3804275,
        "c": "X"
    },
    {
        "a": "B",
        "b": 2317615.875,
        "c": "X"
    },
]

table = Table(schema, index="key")
table.update(data)
table.remove(["A"])
view = table.view(
    group_by=["c"],
    columns=["b", "alias"],
    expressions=['//alias\n"b"'],
)

# This has the wrong value for the `alias` column
view.to_records()

# This however would've worked, `.remove()` after `.view()`
table.remove(["A"])
view.to_records()

Fixes #1710 (???)