Add load_layered_config: load the global ~/.omnigraph/config.yaml layer under the
project ./omnigraph.yaml (or --config) layer and merge them, returning the merged
config, its provenance, and per-layer deprecation warnings. The CLI's load_cli_config
now uses it — so a graph/server/defaults defined only in the global config is usable
from any directory with no project file. The server stays single-layer (a deployment
manifest must not pick up ambient $HOME state).
Global and cwd-default files are optional (absent = no layer); an explicit project
--config still errors if missing. base_dir stays the highest loaded layer's config
dir so relative ad-hoc --query paths resolve as before. The CLI test harness pins
OMNIGRAPH_HOME to an empty temp dir so tests never read the developer's real global
config.
Top-level `policy:`/`queries:` (the single-graph-mode blocks) are removed under
`version: 1` — policy and stored queries belong on the owning `graphs.<name>`
entry. Both are rejected at load (pointing at the per-graph block) and honored-
but-warned under the legacy schema; the per-graph `graphs.<name>.policy`/`queries`
blocks are nested, so the key scan leaves them untouched. Drops the now-illegal
empty top-level `policy: {}` from the shared CLI test helpers.
`defaults:` is the canonical CLI/client-defaults block. The legacy spelling
`cli:` is accepted as a serde alias and honored under the legacy schema, but
rejected under `version: 1` (pointing at the new spelling) and flagged by a
deprecation warning. Generalizes the version-gated key scan into
`legacy_top_level_keys`, which now drives both the v1 rejection and the legacy
warnings via a shared migration-hint table. Renames the config accessors
(cli_* -> default_*) and repoints the CLI call sites; migrates the init
scaffold, the example config, and the shared test helpers to `defaults:`.
Surface the config deprecation notices at load: the CLI prints them via
`eprintln!` in `load_cli_config`; the server logs them via `tracing::warn!` in
`load_server_settings` (a no-op under tests with no subscriber). `omnigraph init`
now scaffolds a `version: 1` config using `storage:` instead of the legacy `uri:`.
Migrate the shared test fixtures (`support::local_yaml_config` →
`version: 1` + `storage:`; `remote_yaml_config` → `version: 1` + `servers:`/`server:`)
to the current schema so they don't trip the new warnings; the resolved `.uri` is
unchanged, so the tests behave identically. The no-`version:` notice is gated on a
loaded config file, so commands with a bare URI and no `omnigraph.yaml` stay quiet
(verified by the existing command-deprecation tests, which still pass). Extends the
`init` test for the scaffold shape and adds a legacy-config-warns test.
mutate_as and load now write directly to target tables and call the
publisher once at the end with per-table expected versions; the Run
state machine, _graph_runs.lance writers, __run__ staging branches,
and server /runs/* endpoints are removed. Multi-statement mutations
remain atomic at the manifest level via an in-memory MutationStaging
accumulator that gives read-your-writes within a query and a single
publish at the end. Concurrent-writer conflicts surface as
ExpectedVersionMismatch (HTTP 409 manifest_conflict) instead of the
old DivergentUpdate merge shape. Documents one known limitation in
docs/runs.md: a multi-statement mid-query failure where op-N writes
a Lance fragment and op-N+1 fails leaves Lance HEAD ahead of the
manifest until a follow-up introduces per-table Lance branches.
Co-Authored-By: Claude Opus 4.7 <noreply@anthropic.com>
