It's somewhat difficult to tell exactly what your situation is. If "test_R1.fastq" is out of sync with "test_R2.fastq", then sync them before proceeding (BBTools has a convenient function for that as I recall).

In any case, you can subsample pairs of fastq files with BBTools as well. That's described here: Select sequences from fastq.gz file

You cannot sample two files and expect the reads to be paired if they have a different number of sequences. As Devon said, you need to pair the reads first. Here is a lightweight solution using Pairfq:

curl -sL git.io/pairfq_lite | perl - makepairs -f test_R1.fastq -r test_R2.fastq -fp test_R1_p.fastq -rp test_R2_p.fastq -fs test_R1_s.fastq -rs test_R2_s.fastq

Now you can sample the paired files and get what you expect.

seqtk sample -s100 test_R1_p.fastq 5 > test_R1_p_5.fastq
seqtk sample -s100 test_R2_p.fastq 5 > test_R2_p_5.fastq

There is inline documentation for above command so ./pairfq_lite will show you the usage, and there is more information about that specific command on the wiki online. If you have a lot of sequences and little memory, I would recommend installing the program from the link above and using the indexing method (if you use this approach).