Populate download cancellation output

[FrogAi]

Summary

• Populate the existing mapdExtendedOut.downloadProgress.cancelled field.
• Map the internal DownloadProgress.Canceled state through the generated serializer.
• Leave download lifecycle, queueing, and cancellation behavior unchanged.

Motivation

The download owner already records cancellation in settings.DownloadProgress.Canceled, and main transfers progress snapshots into ExtendedState. The public MapdDownloadProgress schema defines cancelled, its generated accessor already exists, and the output documentation and CLI use that field.

ExtendedState.setDownloadProgress copies the active state, totals, locations, and per-location details but never copies cancellation. Even when ExtendedState holds a canceled terminal snapshot, the serialized field remains at its default false value.

Implementation

The serializer now calls SetCancelled with s.DownloadProgress.Canceled beside the existing SetActive call.

The internal American spelling and public British spelling remain unchanged; this is the missing mapping between their existing contracts.

Behavior

Snapshot	Base 68813e05	This change
Active and not canceled	active=true, cancelled=false	Unchanged
Terminal and canceled	active=false, cancelled=false	active=false, cancelled=true
Counts, locations, and location details	Populated	Unchanged

Validation

An external Linux oracle constructed the real extended-output struct, invoked ExtendedState.setDownloadProgress, marshaled and unmarshaled the Cap'n Proto message, and read the result through generated accessors.

• Base 68813e05 failed because a terminal snapshot with Canceled=true serialized cancelled=false.
• The candidate preserved active-state semantics and serialized the terminal cancellation state correctly.
• Total/downloaded counts, the location list, and per-location details remained unchanged.
• The focused oracle passed ten repetitions under the race detector.
• go test ./..., go test -race ./..., go vet ./..., and go build ./... passed on Linux/amd64 with Go 1.25.1.
• FrogAi Actions Build #48 ran the branch make build path successfully for exact commit 5688b97.

Compatibility

• The schema field, ordinal, type ID, generated accessors, event union, queue, documentation, and CLI reader already exist and are unchanged.
• Existing consumers that ignore cancelled are unaffected. Consumers that read it receive the state already held by ExtendedState.
• PR Own the map download progress lifecycle #107 separately owns terminal-progress retention and request cancellation. It does not modify extended_state.go, so this serializer correction does not duplicate or conflict with that lifecycle change.
• No download scheduling, HTTP, archive, progress-channel, message-frequency, configuration, exported API, or file-format behavior changes.

Audit follow-up — this line has no observable effect until #107 lands

An independent audit confirmed the added line is correct and sets the right field from the right source, but found that the Behavior table describes the serializer boundary rather than what a fork actually receives on current main.

Canceled=true appears only in the terminal progress snapshot, which is delivered through a non-blocking send into a capacity-1 channel polled once per 50 ms — and that terminal snapshot is virtually always dropped. Measured on current main: 0 of 30 runs published cancelled=true; what is actually published is active=true, cancelled=false. The table holds given such a snapshot, and that was reproduced, but the snapshot does not arrive.

PR #107 is what makes this work: its publishProgress() evicts the stale capacity-1 entry before re-sending, so the terminal canceled snapshot actually reaches ExtendedState. The file-level independence claim is accurate — #107 does not touch extended_state.go — but the dependency is behavioral, so land #107 first or alongside this one; by itself, this publishes a field that stays false. Composition with #106 and with #107 was verified by trial merge, build, and test.