feat(upload): make --artifact optional on upload build

[krassx]

Omitting --artifact registers the build and ships no bytes (request_artifact_upload: false).

Why

• The mode already exists: build::run skips packing and never reads the artefact when the flag is false, and xcode post-action uses it. No CLI caller could reach it — --artifact was required and the flag hard-coded true.
• It is the default path on Android and iOS, which ship bytes only when size analysis is opted into (sizeAnalysis.enabled, BUGSEE_SIZE_ANALYSIS_ENABLED).
• A web build has no single artefact, so it is the only path a web build can express.

Behaviour

invocation	result
upload build --payload-json p	registers; nothing packed or sent
upload build --payload-json p --artifact app.aab	unchanged
--mapping / --chunked / --out without --artifact	exit 20
--artifact /nope	exit 10, as before

--deps/--timings still travel without an artefact — the build-info bundle is a separate upload. The three rejected flags only describe how artefact bytes move; dropping a --mapping silently would cost symbolication.

Tests

6 CLI-level tests asserting both directions of the flag (with an artefact, the same log line says true and the ZIP is still packed), so "ships nothing" cannot pass on a run that never consulted it. They strip tracing's ANSI to assert the field's value rather than its name. 3 mutants caught (pinned true, pinned false, guard bypassed). 1 e2e flow: payload intact, request_artifact_upload: false, zero PUTs.

Gates: fmt, clippy -D warnings (0), all unit tests + suites, e2e_flows.py ALL PASS.

Not here

• format: "web" needs no CLI change — the payload passes through verbatim, so the producer writes it.
• The size-check baseline hardcodes format=ipa; web has no size-check in slice 1, so it is deferred.
• The build uuid derivation for web is blocked on what a build record is keyed by — bugsee/bugsee-javascript#8.

First slice of web-build registration (audit §6). Design: docs/design/web-build-registration.md.

🤖 Generated with Claude Code

[claude]

Code review

Makes --artifact optional on upload build so a caller can register a build and ship the build-info sidecars without shipping artefact bytes — the register-only mode already existed inside build::Params/xcode post-action but had no way to be reached from the command line. The new flags-need---artifact guard (--mapping/--chunked/--out) fails closed with exit 20 rather than silently dropping bytes, build::run's register-only path was already covered by existing tests, and the new CLI-level tests assert both directions (with and without --artifact) so the assertion can't pass on a guard that never looked at the flag. --help text, README, and CHANGELOG were all updated for the new optional flag and its interactions.

I checked validation ordering (the pre-existing mapping/deps/timings existence loop runs before the new "needs --artifact" guard, so a nonexistent --mapping path without --artifact surfaces as exit 10 "does not exist" rather than exit 20 "needs --artifact"), the Path::new("") sentinel for the unset-artifact case (only read when request_artifact_upload is true, confirmed in build::run), the wire shape (no field renamed/added — request_artifact_upload already existed in the POST body, this just varies its value), and MSRV/cross-platform/daemonizing (none of that code is touched). Nothing there rises to a real, confident finding.

Findings: None.

Merge as is.

[claude]

Code review

This PR makes --artifact optional on upload build, threading Option<PathBuf> through to the existing request_artifact_upload flag that build::run and xcode post-action already supported but no CLI path could reach. It adds a validation block rejecting --mapping/--chunked/--out when no artefact is given (exit 20, config_invalid), updates the clap doc comment on artifact in the same change, and adds matching CLI-level tests plus an e2e flow.

I traced the new logic against build::run (artefact is only read/packed when request_artifact_upload is true, so the Path::new("") placeholder for the None case is never touched), the exit-code contract in src/exit_code.rs, and the mock-server e2e additions in scripts/e2e_flows.py — all consistent. The --help text, README, and CHANGELOG were all updated to describe the new optional behavior, satisfying the --help-drift rule in CLAUDE.md.

Findings: None.

One thing I noticed but don't consider blocking: the pre-existing "sidecar file must exist" check (--mapping/--deps/--timings) runs before the new "needs --artifact" check, so passing a nonexistent --mapping path with no --artifact returns exit 10 (input_not_found) rather than exit 20 (config_invalid) — the tests only exercise the case where the mapping file exists. Both are legitimate non-fallback codes and the ordering was already there, so this isn't a contract break, just a minor inconsistency worth a mention if you want the two checks to agree on precedence.

Recommend merging as is.