Gaussian guess - guess negative Gaussian if peak 'looks' negative.

If `average([y value at minimum x], [y value at maximum x]) > (min(y) + max(y))/2`, guess a negative Gaussian instead of a positive one.

## Acceptance criteria

- For a Guassian with tails higher than midpoint of y-values, negative Gaussian is guessed
- For a Gaussian with tails lower than midpoint of y-values, positive Gaussian is guessed

## Notes

- Relevant code is here: https://github.com/ISISComputingGroup/ibex_bluesky_core/blob/5de624a00d4236d36f0d2e606afea9c22892e842/src/ibex_bluesky_core/fitting.py#L159
- When guessing a negative Gaussian:
- We still want to guess the centre & width using centre-of-mass - but this will now be centre-of-mass of inverted data. The easiest way to do this will be to guess the centre & width using `cen, width = _guess_cen_and_width(x, -y)`.
- Background guess is `max(y)`
- Amplitude guess is `min(y) - max(y)`

Start at src/ibex_bluesky_core/fitting.py around line 159 and read the existing Gaussian-guessing logic and _guess_cen_and_width entry point. Implement the negative-versus-positive selection described in the issue, including inverted-data centre and width, then verify that both tail configurations produce the expected Gaussian guesses.

Written by the indexing model from the issue text.