text: Wrap BufferId in a newtype

[osiewicz]

This change is two-fold:

1. We want to use more newtypes, and buffer ID's have been represented as u64's all over the place. It feels like a pretty fundamental identifier.
2. I want to switch the internal type of buffer id to NonZeroU64 to make use of niche optimization (as it should reduce text::Anchor size by 20%).

TODO:

• ensure that prod can make use of this change (as is, I'm not sure if JoinChannelResponse.buffer_id can ever be 0).

Dev tl;dr: decreases size of text::Anchor from 40 to 32 bytes by making Option<BufferId> take up 8 bytes instead of 16.

Release Notes:

• N/A

[osiewicz]

Both SERIAL in PostgreSQL and AUTOINCREMENT in SQLite start off at 1, so we should be safe here. We've checked prod db with @maxbrunsfeld and we don't have a channel buffer with id=0 (that aligns with ids starting off at 1), so this PR should be good to go.

[ConradIrwin]

@osiewicz / @maxbrunsfeld This breaks collaboration between versions of zed before and after the change because older clients do in fact send a buffer id of zero (I think if they're sharing the first opened in a workspace), which causes newer clients to hit an assertion in the workspace and panic.

I'm going to assume we're ok for following between stable and preview to be broken for one file per workspace for one week; leave this in place, and fix the assertion to log an error instead.