Personally, I do not think this is a matter of concern. That might be due to the shortening of reads after trimming, which alters the GC% overall content. It's not that trimming reads introduce more GCs. Those are just warnings and your GC% content for all samples still roughly follows a normal distribution. I would proceed with the trimmed samples without worrying too much, this will most likely have no effect on downstream data processing.

Have you already taken a look at those reads? Something like

seqkit grep -s -p '".*[CG]{10,70}.*[CG]{5,70}.*[CG]{10,70}"' -r < your.fastq > CGrich.fastq

should extract at least some of those very CG-rich reads?

From which organism these reads come from?