Using fixed-effects for each person helps model such heterogeneity, but still nothing comes out.

Feddersen et al model very particular heterogeneity, to wit the usual demographic and econometric data:

These include age and its square, the number of household dependents aged between 0 and 24 years, and the natural logarithm of nominal household disposable income for the previous financial year in Australian dollars. Dummy variables are also included for disability status, employment status, marital status and education. These controls are typically the most important determinants of self-reported life satisfaction

That's not going to help ferret out idiosyncratic reaction to weather.

It's just that the old-wives-tales are wrong yet again.

I don't have any old wives to listen to :-) The source is my personal experience and the experience of people I know who don't seem to have any reason to lie about it.

If you really believe that weather relationships could be so obscure and complicated and individual that there are no meaningful average effects

I didn't say that. I said that not seeing average effects does not rule out individual-level effects and that at the anecdotal level I do see these effects.

And there is no reason to generalise to everything. I am sure that "bad air" which is bad enough to produce consistent measurable average effects exists and is not very hard to construct or find.