1. Stop auto-discarding abuse reports which aren't from the target of an attack. Right now, Twitter won't investigate a credible death threat unless you say you're the target.The "We only review…" message is bad UX, after someone has already gone through multiple steps, but it doesn't necessarily mean the report has been auto-discarded. The fact that their form lets you submit them at all suggests it might be used as a secondary signal – just one that doesn't guarantee a formal 'review' and response. And Twitter might be offering this partial-but-disappointing path to collect data for some other future opening to third-party flagging.
Still, third-party flagging brings its own problems. If it triggers anything automatic, it gets abused for other group-persecution/censorship purposes. And if it always requires staff review, it'll waste a lot of staff time on harmless joking between friends. So I don't blame Twitter for currently setting a 'bright line' bound on what level of seriousness (target reporting) triggers staff involvement.
2. Limit new accounts ability to send messages to people who don't follow them
This is a plausible idea and I wouldn't be surprised to see some form of it soon, as a user-enabled option.
Note, though, that it still disadvantages good-faith new-users, as well: the person who was moved to register to message someone would now have a frustrating, "you're-not-good-enough-to-message-them" experience. How many hundreds or even thousands of good-faith interactions would you be willing to prohibit, to prevent a few episodes of nastiness?
3. Apply blocks to email addresses, requiring griefers to churn validated addresses rather than reusing them
This seems reasonable. Are you sure Twitter isn't already doing this?
4. Give you better filters for notifications: i.e. currently there's no way to have it default to only notifications from people you follow so you can't avoid seeing crap unless you use a bookmark. Similarly, filtering out messages from new accounts you don't follow or which don't have n followers in common, etc. etc. would reduce the dog-pile effect.
Another reasonable idea I wouldn't be surprised to see soon.
5. Allow friends to help: auto-block/mute accounts which people you follow have blocked (specific people, threshold, etc.); hide tweets which people you follow have flagged as annoying / abusive; etc. There's a ton of interesting opportunity to use the social network for good here.
Also reasonable, if all the pesky details can be hammered out. Note some of these features are being enabled by 3rd-party apps like 'Block Together' (blocktogether.org). I'd expect the best ideas to eventually work their way into the official Twitter offering.
But it's actually quite hard for Twitter to rapidly innovate in this regard, because users are very ornery about slight changes in such options. Case in point: last December's attempt to change 'block' to be more of a 'mute'. Every tiny change in moderation capabilities triggers controversy, and even things that don't work very well, but become loved as a security blanket, will be hard to roll-back once enabled. Hence the lack of any big, crowd-pleasing moves by the platform proprietor.
I definitely agree with you that frictional costs are important: incentives matter, and driving up the costs and dampening the rewards will help minimize (though not abolish) bad behavior.
I'm unsure if wisdom from spam battles will be applicable, though. A determined and deranged abuser isn't playing a bulk numbers game, like a spammer, who is just hoping some profitable percentage of spam sneaks through and 'converts'. The abuser is instead focusing on few targets, more like a spear-phisher, and hand-crafting all the abuse. So quite different tactics may be required.