Overwrite Read sequences in bam with reads from Fastq

I had to modify my original reads (masking few nucleotides), now my alignment went well (i use bowtie2) but I need my original reads in my sam. So I need to catch the original read from my fastq to reinject it into my bam (I know the cigar would be compromised but I don’t care).

What would be the most efficient?

My idea would be to use python to parse the fastq and the bam with Python (using pyfastx and pysam), but I’m wondering if something more efficient already exist.