docs: rename runs.md/runs.rs → writes and repoint all references (#131)

The Run state machine was removed in MR-771 (v0.4.0); `docs/dev/runs.md`
and `crates/omnigraph/tests/runs.rs` have since documented and tested the
direct-publish write path, so the "runs" name was misleading.

- git mv docs/dev/runs.md → docs/dev/writes.md (reframe H1 + intro;
  keep MR-771 history note)
- git mv crates/omnigraph/tests/runs.rs → tests/writes.rs (reframe header)
- repoint every runs.md / runs.rs reference across docs, AGENTS.md, and
  source comments
- fix four pre-existing broken `docs/runs.md` links (the file never lived
  at that path) to `docs/dev/writes.md`
- fix the stale v0.4.0 anchor to the live section

No behavior change: every source edit is a comment. Engine builds and the
renamed test passes 25/25; scripts/check-agents-md.sh passes.

The run-removal cleanup itself (run_registry.rs guard, __run__ prefix) is
deferred to MR-770.

crates/omnigraph-cli/tests/system_local.rs

@@ -991,7 +991,7 @@ query vector_search($q: String) {
 
 // The publisher CAS conflict shape is verified end-to-end at the engine
 // level in
-// `crates/omnigraph/tests/runs.rs::concurrent_writers_one_succeeds_one_gets_expected_version_mismatch`
+// `crates/omnigraph/tests/writes.rs::concurrent_writers_one_succeeds_one_gets_expected_version_mismatch`
 // and at the HTTP boundary in
 // `crates/omnigraph-server/tests/server.rs::change_conflict_returns_manifest_conflict_409`.
 // A CLI-level race would be timing-dependent; with direct-publish the

crates/omnigraph/src/db/manifest/recovery.rs

@@ -2,7 +2,7 @@
 //!
 //! This module implements the building blocks of the per-sidecar recovery
 //! sweep that closes the documented Phase B → Phase C residual (see
-//! `docs/dev/runs.md` "Open-time recovery sweep"). The high-level shape:
+//! `docs/dev/writes.md` "Open-time recovery sweep"). The high-level shape:
 //!
 //! 1. Each writer that performs a multi-table commit writes a small JSON
 //!    sidecar at `__recovery/{ulid}.json` BEFORE its per-table

crates/omnigraph/src/loader/mod.rs

@@ -613,7 +613,7 @@ async fn load_jsonl_reader<R: BufRead>(
     } else {
         // LoadMode::Overwrite keeps the legacy inline-commit path —
         // truncate-then-append doesn't fit the staged shape (see
-        // `docs/runs.md` "LoadMode::Overwrite residual"). The recovery
+        // `docs/dev/writes.md` "LoadMode::Overwrite residual"). The recovery
         // sidecar is not applicable here because the writer doesn't go
         // through MutationStaging; per-table inline commits + a final
         // manifest publish handle their own residual via the documented

crates/omnigraph/src/table_store.rs

@@ -49,7 +49,7 @@ pub struct DeleteState {
 /// `exec/mutation.rs`) and the bulk loader (`loader/mod.rs`). The
 /// intent: defer Lance commits to end-of-query so a mid-query failure
 /// leaves the touched table at the pre-mutation HEAD instead of
-/// drifting ahead. See `docs/runs.md` for the publisher-CAS contract
+/// drifting ahead. See `docs/dev/writes.md` for the publisher-CAS contract
 /// this builds on.
 ///
 /// `transaction` is opaque from our side — Lance owns its semantics. We
@@ -901,7 +901,7 @@ impl TableStore {
     /// Lift path: either a Lance API extension that lets
     /// `MergeInsertBuilder` accept additional staged fragments, or an
     /// in-memory pre-merge here that folds prior staged batches into the
-    /// input stream. See `docs/runs.md`.
+    /// input stream. See `docs/dev/writes.md`.
     pub async fn stage_merge_insert(
         &self,
         ds: Dataset,

crates/omnigraph/tests/staged_writes.rs

@@ -2,7 +2,7 @@
 //! exercise `stage_append`, `stage_merge_insert`, `scan_with_staged`,
 //! and `count_rows_with_staged` directly against a Lance dataset — no
 //! Omnigraph engine involved. The engine-level use of these primitives
-//! is exercised by `tests/runs.rs`.
+//! is exercised by `tests/writes.rs`.
 //!
 //! Test surface here:
 //! 1. `stage_append` + `scan_with_staged` shows committed + staged data
@@ -709,7 +709,7 @@ async fn stage_create_inverted_index_does_not_advance_head_until_commit() {
 ///
 /// **When Lance #6658 lands**: this test will need to flip — replace
 /// the assertion with a `stage_delete` + `commit_staged` round-trip
-/// and remove the residual line in `docs/runs.md`.
+/// and remove the residual line in `docs/dev/writes.md`.
 #[tokio::test]
 async fn delete_where_advances_head_inline_documents_residual() {
     let dir = tempfile::tempdir().unwrap();

crates/omnigraph/tests/writes.rs

@@ -1,7 +1,7 @@
-//! Tests for the direct-to-target write path (Run state machine
-//! removed). The Run/`__run__` staging branch / RunRecord state machine no
-//! longer exists; mutations and loads write directly to target tables and
-//! commit once via the publisher's `expected_table_versions` CAS.
+//! Tests for the direct-publish write path: mutations and loads write
+//! directly to target tables and commit once via the publisher's
+//! `expected_table_versions` CAS. (History: this replaced the removed Run
+//! state machine / `__run__` staging branches / RunRecord — MR-771.)
 //!
 //! What this file covers:
 //! - No `__run__*` branches are created by load or mutate.