Different samples are sequenced with different sequencing platforms, such as Illumina Miseq, Illumina Novaseq, Illumina Hiseq.

These are all Illumina sequencers so the basic technology is identical (though the chemistry may be different). You should be able to use standard Illumina QC and analysis techniques for all of these data.

I have collected the public whole-genome sequencing raw read data from NCBI.

and

I am working with bacteria whole-genome sequencing data and planning to search SNPs.

Collecting public sequence data and trying to do meta-analysis to identify SNP's sounds vague. What do you actually hope to get from this? Are you focusing on sequence from a particular species/strain?

The simplest thing to do to harmonize all your different reads is to trim them all to the same length. You want the tricky repetitive regions to be equally difficult in all samples.

The chemistry differences between the different platforms are very small, so I wouldn't worry about that.

Then, should I set different minimum length? Because 50bp from the sample of average read length 75bp and 300bp should imply different meanings. Previous sample wouldn't be a big problem, but the latter one would mean severe quality problem, doesn't it?

In this case, I would suggest that you set a standard on what read-Lenght you want to consider. say you set 150bpx2 as minimum, then you can put all the data through 1 standard pipeline, with similar flags for each tool. This will be helpfull in dealing with batch-effect (if that can happen in WGS data too)