seq: optimize slow path loop by implementing AddAssign for ExtendedBigDecimal

[CrazyRoka]

This PR optimizes the "slow path" of seq by replacing the allocation-heavy addition loop with in-place mutation.

Currently, the print_seq loop performs the following operation:

value = value + increment.clone();

It explicitly clones the increment (a potential BigInt allocation) on every iteration.

I have implemented std::ops::AddAssign<&ExtendedBigDecimal> for ExtendedBigDecimal and updated seq to use:

value += &increment;

I reviewed the upstream bigdecimal crate and noticed that addassign_bigdecimal_ref currently contains a TODO and performs a to_owned() internally. Even with that upstream limitation, this PR provides immediate benefits.

[sylvestre]

What happened? :)

[CrazyRoka]

I expected CodSpeed to report performance improvement, but that didn't happen.

I think existing benchmarks are not correctly exercising the slow path:

coreutils/src/uu/seq/benches/seq_bench.rs

Lines 18 to 24 in 8cf88cd

	/// Benchmark large integer
	#[divan::bench]
	fn seq_large_integers(bencher: Bencher) {
	bencher.bench(|| {
	black_box(run_util_function(uumain, &["4e10003", "4e10003"]));
	});
	}

Apart from that, improvement might be minimal, because bigdecimal is cloning as well.