Add experimental LSP-based context retrieval system for edit prediction (#44036)

To do

* [x] Default to no context retrieval. Allow opting in to LSP-based
retrieval via a setting (for users in `zeta2` feature flag)
* [x] Feed this context to models when enabled
* [x] Make the zeta2 context view work well with LSP retrieval
* [x] Add a UI for the setting (for feature-flagged users)
* [x] Ensure Zeta CLI `context` command is usable

---

* [ ] Filter out LSP definitions that are too large / entire files (e.g.
modules)
* [ ] Introduce timeouts
* [ ] Test with other LSPs
* [ ] Figure out hangs

Release Notes:

- N/A

---------

Co-authored-by: Ben Kunkle <ben@zed.dev>
Co-authored-by: Agus Zubiaga <agus@zed.dev>

crates/text/src/anchor.rs

@@ -8,10 +8,14 @@ use sum_tree::{Bias, Dimensions};
 /// A timestamped position in a buffer
 #[derive(Copy, Clone, Eq, PartialEq, Hash)]
 pub struct Anchor {
+    /// The timestamp of the operation that inserted the text
+    /// in which this anchor is located.
     pub timestamp: clock::Lamport,
-    /// The byte offset in the buffer
+    /// The byte offset into the text inserted in the operation
+    /// at `timestamp`.
     pub offset: usize,
-    /// Describes which character the anchor is biased towards
+    /// Whether this anchor stays attached to the character *before* or *after*
+    /// the offset.
     pub bias: Bias,
     pub buffer_id: Option<BufferId>,
 }