Sensitive to `COMMODITY_BALANCE_EPSILON_FOR_CANDIDATES`

Introduced in #1387, I've found that this parameter only really works as intended over a less than 2-orders of magnitude range:
- too small and it has not impact
- too big and candidate assets cannot meet the demand

If this is reliable, then maybe that's ok, but I've only tested this with the limited example models in the repo. I also haven't investigate the relationship with the related `candidate_asset_capacity` parameter. I imagine the relative value of this is important, so it may be better to define the epsilon parameter relative to this, or just hardcode them both.

I also suggested an alternative idea [here](https://github.com/EnergySystemsModellingLab/MUSE2/issues/1386#issuecomment-4864876439), which would avoid having to do the dispatch with candidates all together.

Start by reproducing the behavior of COMMODITY_BALANCE_EPSILON_FOR_CANDIDATES with the limited example models in the repository. Compare its effect with candidate_asset_capacity across the tested ranges, and review the alternative dispatch approach in issue #1386. Done means the parameter relationship and expected behavior are established clearly enough to guide a focused change.

Written by the indexing model from the issue text.