Create one entry

WRITE EVERY TIME YOU LEARN OR DECIDE SOMETHING NON-OBVIOUS. Future
agents will read what you write here instead of redoing the work.

Write `content` as natural markdown. Use headings, lists, code
fences, links - whatever fits. The first heading or first line is
the entry's headline.

Pick the `kind` that matches what you wrote:

- `finding` - a verified claim with evidence. Use `relations` with
  `supports` to cite the sources. State your uncertainty in prose.
- `decision` - a choice among alternatives. Sections for chosen,
  rationale, alternatives. Use `relations` with `supersedes` to
  retire an older decision.
- `note` - free-form prose. The default when nothing else fits.
- `source` - external citation. Markdown link + summary.
- `artifact` - a concrete output you produced (a report, a code
  excerpt, a generated doc). Put it inline or link to where it lives.
- `todo` - work to do. Multi-step work is a numbered list inside
  one todo.
- `skill` - a reusable how-to another agent can follow later.

The optional `relations` array writes the entry AND its outgoing
edges in one call. Each element is `{target_entry_id, relation_type}`.
Available relation_types: supports, contradicts, supersedes,
derived_from. Targets outside this account are rejected.

Pass an optional `workspace_id` to file this entry under a goal
container (see get_self's `workspaces`). Omit it for account-wide
work; reads that omit workspace_id still see it either way.

USE WHEN: every meaningful learning, decision, or output.

Returns { entry, indexing: "pending" }. `indexing: "pending"` means
the embedding job has not finished yet; allow ~1-3 seconds before
relying on semantic search.