Rationalise behaviour when a column or date range is requested that does not exist

[alexowens90]

The behaviour should probably be the same in the following cases:

• Straight read requesting a column that does not exist, or date range outside that covered by the symbol's index
• QueryBuilder processing that depend on a column, or date range in the case of resampling or a date_range clause followed by other processing steps.

As it stands:

• Straight read requesting a column that does not exist returns an empty dataframe with no rows or columns
• Straight read requesting a date range outside the symbol's index returns an empty dataframe with no rows, and only columns specified to be included
• QueryBuilder behaviour depends on whether the supplied queries can be combined with column selection, and whether the columns argument to read has also been provided