50,000 contractors may have clearances, but they don't all have access to those records. Remember, the whole point of Snowden going in with
sysadmin access, and then stealing credentials and otherwise impersonating 20 other people, is precisely
because the stuff Snowden wanted to leak was so compartmentalized.
Your 1) has already happened, AFAIK. Perhaps we need to have different (i.e. more active and intrusive) Congressmen on the subcommittee so that they don't act so shocked when details come out, but that's a different issue.
Another key note is that unlike almost every other subcommittee, these Congressional overseers don't get the ability for political patronage out of what they do. The Chairman of the House Armed Services Committee, for instance, can be greatly influential in drawing massive DoD dollars to their district, but there's not much the Intelligence committees can do in the same vein, aside from random data centers perhaps.
Even then, I know there was a determination by Sen. Feinstein earlier that EO 12333 collection was not in their oversight bailiwick. That wasn't NSA refusing to discuss it, it was the subcommittee deciding it was out of their jurisdiction.
There are many Federal agencies and groups that do monitor EO 12333 (though it is certainly unclear what type of oversight they are providing across the totality of programs).
Beyond all that, I do agree that nothing going on within NSA should be out-of-scope for oversight (including Congressional oversight, if it comes to it). From the nuclear propulsion perspective, we had routine audits of essentially every major program, institutionalized. I don't know how the Congressional subcommittees do their business but it would be a good idea to institute the same there, and also ensure that NSA/IC staff (contractors too) can always feel comfortable coming to a staffer from that Congressional subcommittee with details of abuses.
I'd give the NSA some credit on security through obscurity because you're talking about intelligence programs, not computer code or algorithms. A rule of thumb from crypto research does not necessarily directly apply to anything with the word "communications" in it. ;)
In fact the only real problem I see in 3) is that (assuming you're talking about public discussion) the people who most need to care (the American public) are going to have their eyes glaze over immediately, the people who might actually be able to craft countermeasures would be intensely interested, and it wouldn't even prevent government abuse; how easy is it to craft a good system of access controls for public approval but then add workarounds as needed in practice?