DimensionInterval breaks when `intervals` shares a `dim` with another `coord`

## Context

A crucial requirement for the Interval Index is the ability to select by `word` not just intervals. For example in this case:

A simple approach is to create a custom index that keeps `word` and `intervals` in sync. However, that adds complexity and makes this less general and reusable by other use cases. Instead, we can swap the dimensions a bit. Instead of having `intervals` have `dim=intervals` we set `word` as the dimension coord.

This allows you to `sel` or `isel` on `word` which is the most meaningful space for these users to be selecting over.

## Problem

I'm struggling to get `sel/isel` to work in this case when I select based on either `intervals` or `time`. Seemingly because our indexer isn't able to convey that `word` also needs to be sliced at the same time as `intervals`.

Using the built in example dataset:

```python
from linked_indices.interval_index import DimensionInterval
from linked_indices.util import interval_dataset
ds = interval_dataset(interval_dim="word")
ds = ds.drop_indexes(["time"]).set_xindex(
["time", "intervals"],
DimensionInterval,
)
ds
```

doing

`ds.isel(time=5)`

gives

`ValueError: conflicting sizes for dimension 'word': length 1 on 'intervals' and length 4 on {'word': 'word', 'time': 'time'}`

but we can fake it by manually slicing word and everything works.

`ds.isel(time=5, word=0)`

we need to find a way to have that happen automatiaclly. I think it could be managed for `sel`, but I don't think that `isel`'s return value is not expressive enough for the index to handle this on it's own.

Similarly

`ds.sel(time=5)`

gives

`ValueError: dimension 'word' already exists as a scalar variable`

and

`ds.sel(time=slice(1,20))`

`ValueError: conflicting sizes for dimension 'word': length 1 on 'intervals' and length 4 on {'word': 'word', 'time': 'time'}`

Start with linked_indices.interval_index.DimensionInterval and reproduce the reported cases using linked_indices.util.interval_dataset(interval_dim="word"). Run ds.isel(time=5), ds.sel(time=5), and ds.sel(time=slice(1,20)) after setting the x-indexes as shown. Done means selecting by time or intervals automatically keeps word and intervals aligned without the reported dimension conflicts.

Written by the indexing model from the issue text.