This strategy is based on standard nix tools. Get the *first two sequences:

awk -v RS='>' 'NR>1 { gsub("\n", ";", $0); sub(";$", "", $0); print ">"$0 }' seq.fa \
    | head -n 2 \
    | tr ',' '\n'

It assumes the semicolon ; doesn't occur in the sequence names.

Replace head with tail to get the last sequences.

For the first 100k sequences:

reformat.sh in=data.fasta out=100k.fasta reads=100000

You can also get a random 100k sequences with Reformat, but not the last 100k.

Assuming no linebreaks in sequences, i.e every record is exactly two lines. I believe for fastq files the multiplier would be 4:

head -n 200000 input > output

Alternatives: seqkit head and seqkit range:

(1) Leading 100000 records:

seqkit head -n 100000 input.fa
seqkit range -r 1:100000 input.fa

(2) Last 100000 records:

seqkit range -r -100000:-1 input.fa

(3) Other ranges:

seqkit range -r 100001:200000 input.fa

REQUESTED_LINES=10
awk "/^>/ {n++} n>$REQUESTED_LINES {exit} {print}" input.fasta > output.fasta

hdl = gzip.open(file, 'rt')
records = SeqIO.parse(hdl, 'fastq')
first_read = next(records)
printfirst_read.id)