I can say with a healthy dose of confidence that there will not be 50-100k experimentally characterized promoters, at least not directly characterized. Maybe you get to those numbers if you use the homology to already characterized promoters, but I doubt that as well.

The process was very slow 20-30 years ago when people used to study promoters in detail, so I'd estimate the number of characterized promoters from those days to be in couple of thousands at most. It has picked up since then based on indirect approaches, but most people today don't study things at such a "local" level.

I think your best bet is to study poly-cystronic transcripts from microarray or RNAseq data, as those have to come from promoters. Likewise, there are many highly confident promoter predictions that drive poly-cystronic transcripts. The logic says that they almost have to be accurate when gene spacers are shorter than 20bp. Lastly, the coding density in prokaryotes is 85-90%, so all intergenic regions longer than 40-50 bp are more likely to be promoters than not. I think with some filtering for common upstream motifs and TF binding elements, there should not be too many false positives among predicted promoters.