Minor comment, but that's fairly dependent on the field. Every morning I am at work I take ~5–10 minutes to skim the titles of the day's new arXiv/astro-ph postings[0] to see if there's anything interesting. If there is, I'll mark it for later, save it, or spend some time reading/skimming it. That's fairly common among my peers in the field. There are even websites that piggyback on arXiv to facilitate voting on papers to create agendas for discussion groups or journal clubs[e.g., 1].
No university affiliation is required, but you are required to 'belong to the scientific community', and to prove that you will need someone to vouch for you and your work.
It's actually a higher bar of entry for arXiv than it is conferences and journals!
1. A strong LOR from a PI managing a known lab(but with few moderate pubs on my name) vs. 2. No great LORs but self-publishing one or two good papers(quantified with good number of citations and overall quality)
Why do you see these things as mutually exclusive? You could self-publish and use it as an "in" to working for a PI over the summer. If you do well and the PI likes you, they likely have a lot of sway with admissions and can help you get into whatever grad program they work with.
Disclaimer: Never applied/went to grad school but know many people who did... I've seen the strategy I described work at a high tier school.
This. Publishing is not an end onto itself. Think of it as a chance to get your foot in the door of the research community.
I myself don't put much faith in LoRs unless they're from someone I know pretty well. Other people differ.
If you're sure you can produce a good paper, I think that's better than a LoR. But research can be hit-or-miss.
However, it's hard to write well for an audience in a given field without having gone to grad school in that field. Readers will sense that your paper is from an outsider. So stick to topics that people will believe an outsider about. Proofs and working systems are good, criticisms of common methodologies are bad.
When you publish, especially on arXiv, you still have to get people to read it. Zero people wake up in the morning thinking "I'll just browse for new papers from people I've never heard of." You might email a few authors of recent papers you're citing saying, "Hey, I really liked your paper on ___. I've just published something related on arXiv that addresses a problem you mention in your future work section..."