fix(cli): keep identifier fields intact in list projection shortening

[ysyneu]

Defect

The compact list projections (incident list, incident similar, alert-event list in --output-format json|toon) bound their output to the 16 KiB list budget by applying one "fair" per-field byte cap to every string value. When a page overflows badly enough that the fair cap lands below an identifier's own length, identifier fields — incident_id, alert_id, event_id, alert_key, … — are clipped mid-value with a trailing ....

User-visible symptom: a consumer that filters the JSON output by exact match (jq 'select(.alert_key == "...")') or passes an id back into a follow-up detail call gets nothing — a clipped identifier matches no real value, and the empty result is indistinguishable from "nothing matched". Whether an id survives depends on what else happened to overflow on that page, so exact-match filtering over list output is unreliable by default.

Root cause

boundProjectedList (internal/cli/fieldproject.go) computed maxLen, ran its fits() trials, and applied the final cap uniformly over all string values — no distinction between identifiers (values a consumer matches, filters, or passes back verbatim) and free text (values a human reads).

Fix

Add isIdentifierField(key) (keys ending in _id or _key) and exempt those fields at all three points of boundProjectedList:

1. the maxLen sizing scan skips identifier keys;
2. fits() copies identifier values verbatim into the trial rows;
3. the apply loop skips identifier keys before total++, so the shortening note's denominator counts only shortenable strings.

Identifiers now survive byte-intact — or, when a page's irreducible content (identifier strings + keys + envelope) alone exceeds the budget, the existing narrowing error fires (request fewer rows (--limit) or fewer --fields) instead of shipping clipped ids. Note format and error text are unchanged; boundProjectedDetail, all projection call sites, flag help, and skill docs are untouched.

Tests

• TestBoundProjectedListNeverShortensIdentifierFields (json + toon subtests): 10 rows of {event_id, alert_key, title(500B)} at a 1400-byte budget — both id fields byte-identical in every row, every title shortened with the ... marker, the note names title but not the identifier fields, and the re-marshaled output fits the budget. Against the old implementation the json subtest fails: the fair cap lands at 27 bytes and clips the 32-byte alert_key.
• TestBoundProjectedListIdentifierOnlyOverflowErrors: 12 rows of {incident_id} at a 512-byte budget — errors with "request fewer rows" and leaves every row unmutated. Against the old implementation it fails: the old code clipped every 24-byte id to 12 bytes and returned success.
• TestBoundProjectedListNeverEmitsUnmarkedTruncation (pre-existing): its 100-row fixture only ever fit because the old code clipped the ids to fit the budget — under the exemption that page errors by design, so the fixture is resized to 90 rows, keeping it in the "titles shorten, ids intact" regime the test is about. This is the intended contract: a page whose irreducible identifier content alone exceeds the budget now fails with the narrowing error instead of shipping mangled ids.

Verification

$ go test ./internal/cli/ -run 'Projection|BoundProjected'
ok  	github.com/flashcatcloud/flashduty-cli/internal/cli	0.808s

$ go test -race ./internal/cli/
ok  	github.com/flashcatcloud/flashduty-cli/internal/cli	5.758s

$ make check   # gofmt -s + gci, golangci-lint, go test -race ./..., go build
bin/golangci-lint run --allow-serial-runners
0 issues.
go test -race ./...
ok  	github.com/flashcatcloud/flashduty-cli/cmd/flashduty	6.258s
ok  	github.com/flashcatcloud/flashduty-cli/internal/cli	(cached)
ok  	github.com/flashcatcloud/flashduty-cli/internal/cmd/cligen	1.566s
ok  	github.com/flashcatcloud/flashduty-cli/internal/cmd/skilldoc	2.963s
ok  	github.com/flashcatcloud/flashduty-cli/internal/config	1.516s
ok  	github.com/flashcatcloud/flashduty-cli/internal/output	2.334s
ok  	github.com/flashcatcloud/flashduty-cli/internal/skilldoc	2.663s
ok  	github.com/flashcatcloud/flashduty-cli/internal/timeutil	1.776s
ok  	github.com/flashcatcloud/flashduty-cli/internal/update	2.937s
go build -v ... -o bin/flashduty ./cmd/flashduty