Flags aren't meant to be unique, they indicate only information like alignment orientation. For your example, you have single-end data so you probably only have 3 different flags (forward, reverse, and unmapped). What you probably meant to do was count the unique number of reads in the file samtools view -c -F 2432 SRR5315260.bam, which will do that (the 2432 will ensure that mates/multimappers/etc. aren't counted more than once). That will presumably be the same number of entries as there are in the fastq file. If not, then you've done something wrong or done some filtering somewhere.

To check the numbers between fastq and sam, you have to consider several factors. At least two come to mind: you might have multimapped reads, so you should have more lines in sam file than reads in fastq. On the other hand, you should check whether or not STAR by default outputs unmapped reads. But overall, if this analysis is not really important for you, I don't think it worth spending time on it, STAR is quite mature software. Anyway, one easy thing you can do is to run fastqc on your fastq file (this will give if number of reads), and then check number of input reads in STAR Log.final.out.

On your commands: samtools view SRR5315260.Aligned.out.sam | cut -f2 | sort | uniq -c - in context of your analysis this doesn't make sense to me. I don't think that each rad has its own unique flag.

wc -l - if you do this on sam file directly it will also count headers.

Another way to count unique reads is to count how many read names appear (This assumes that the first and second reads have identical names, otherwise, do "samtools view -f 64" to only get read 1)

samtools view  SRR5315260.Aligned.out.bam | cut -f 1 | sort | uniq -c | wc -l

And as others have said, counting flags makes no sense.