Fix duplicated UPRN column after reweighting

[crispy-wonton]

Reweighting outputs 2 columns: UPRN and UPRN_right.

It seems to come from the reweighting pipeline because it already exists in the dataset that we load in at the beginning of run_add_features.py. I think it originates from the outer join we use to join the weights to the epc_df. This is an expected polars behaviour that I forgot to account for.

This means, rows which have nulls in UPRN_right should be missing weights. They will either be missing weights because:

• They are in an LSOA which was skipped because we are missing census data for it (mainly affects Scotland, but also England and Wales in some instances)
• We had to drop the row for weighting because the row had a category which was not found in the target data for that LSOA. E.g. if the property type of the row is 'flat' but the census has 0% flats for that LSOA, the row will be dropped.
• They are in Scotland. This dataset was run before reweighting for Scotland was added, so all Scottish rows should be missing a weight.

Here is the count of rows with missing UPRN_right for each country:

>>> epc.filter(pl.col("UPRN_right").is_null())["COUNTRY"].value_counts()
shape: (3, 2)
┌──────────┬─────────┐
│ COUNTRY  ┆ count   │
│ ---      ┆ ---     │
│ cat      ┆ u32     │
╞══════════╪═════════╡
│ Wales    ┆ 38285   │
│ England  ┆ 788475  │
│ Scotland ┆ 1473612 │

All of them have null weights, as expected:

>>> epc.filter(pl.col("UPRN_right").is_null())["weight"].value_counts()
shape: (1, 2)
┌────────┬─────────┐
│ weight ┆ count   │
│ ---    ┆ ---     │
│ f64    ┆ u32     │
╞════════╪═════════╡
│ null   ┆ 2300372 │

Originally posted by @crispy-wonton in #70 (comment)