feat(output)!: print a default field set on -j listings, with --fields all for the full record

[ryanlewis]

Behaviour change

-j on a task listing (every view, filtered list), search and projects no longer prints every key. Without --fields it prints a default set per kind; --fields all prints the full record, byte-identical to the old -j output. A script that reads notes, projectUUID, areaUUID, headingUUID, creationDate, index, todayIndex or trashed off a task listing, or startBucket off a projects row, needs --fields all (or things show <uuid> -j). show -j is unchanged.

Default sets

Defined next to TaskFields/ProjectFields in internal/output/fields.go, in record order.

• Tasks: uuid, title, type, status, start, startBucket, startDate, reminderTime, deadline, stopDate, projectTitle, projectTrashed, areaTitle, headingTitle, tags, repeating, checklistProgress
• Projects: uuid, title, status, start, startDate, deadline, areaTitle, tags, taskCount, openCount

Left out of tasks:

• notes: the bulk of the output (about half of a full listing); an agent reads them for the one item it acts on, with show.
• projectUUID, areaUUID, headingUUID: an agent acts on the item's own uuid; it filters by project or area with --project/--area, which take titles.
• creationDate: rarely used to choose an item; available with --fields.
• index, todayIndex: Things' internal sort keys; the listing is already in the app's order.
• trashed: always false outside the trash view, always true inside it.

startBucket stays in: it is how today -j tells the app's This Evening rows apart, and matching what Things.app shows is the project's goal.

projectTrashed stays in: it is omitempty, so it costs nothing on ordinary rows, and on --project <trashed project> it tells an agent the row will be refused before it tries a write.

Left out of projects: startBucket (a project's This Evening placement only matters inside a today listing, which carries it) and areaUUID (as above). taskCount stays in, because the docs' own "projects with nothing left open" recipe uses it.

all

--fields all on its own prints the full record. all with other names (all,notes) is refused with a plain error, since all already names them. No JSON key is named all, and a test keeps it that way.

Small fixes

• The unknown-field and empty-list messages now end with ", or all for the full record".
• An empty --fields ("", " , ") now gets the unknown field payload with valid and no unknown, instead of the generic error token.
• docs/content/commands.md hard-codes the key lists, so a new test (internal/output/docs_test.go) ties them, and the default sets in commands.md and SKILL.md, to the variables.

Sizes

On a copy of today's Things backup DB, old -j vs new -j (bytes):

listing	before	after
anytime -a "Personal Projects" (94 tasks)	90150	28583
today	53979	16542
upcoming	23020	8030
deadlines	8774	3462
search the	851978	270356
logbook	778483	355580
projects	5857	4577

--fields all matched the old -j byte for byte on every listing above (also inbox, someday, trash, repeating, projects --completed, list --project). Wall time for logbook -j is unchanged (~25 ms).

Docs

SKILL.md, docs/content/commands.md, docs/content/agents.md and the README output paragraph are updated. docs/content/changelog.md is generated from release notes, so it is not edited.