update comments

[unalmis]

Adds some comments to objectives

[rahulgaur104]

Full optimization performance benchmarks were never performed. For example, when optimizing an equilibrium, the new bounce integral from ku/bounce (merged) took almost exactly the same time as the old bounce integration routines. So compute maybe faster but optimization definitely was not. Maybe sparse pullback will change that, maybe not.

I don't actually care at this point, so approving this PR.

[unalmis]

@rahulgaur104 's comment purpose is not clear to me, nor why he feels the need to advertise he does not care. To reduce spread of misinformation for others who are reading, note that ku/bounce (link: #1919 ) is a branch that primarily focused on correcting bugs in DESC and improving convergence; not performance. Still, the performance benefit from that PR is significant if one benchmarks it properly - e.g. compare how much faster it is when a low resolution solver is used (because the convergence improvements make the new low resolution solver as accurate as the old high resolution solver). All of this is shown in detail in #1919 .

Indeed, ku/sparse_pullback (link: #2170 ) and the following PRs reduces memory and improves performance by a factor of the number of flux surfaces the optimization is done with. This was benchmarked and proven (in math and in practice), so rahul comments are unaware of that. One should not diminish the value of my work, which is actually the largest performance boost in terms of memory+compute in desc to date. One performance benchmark demonstrating this (among many others that were shared with visualization) is available here (https://github.com/unalmis/DESC/blob/master/publications/unalmis2025/effective_ripple_profile.py).