SUBS-025: submodule GET recursion leaks raw fetch progress instead of hunks

Input

Context

be get recursing into a submodule prints the sub's fetch progress as RAW terminal text, interleaved with the parent's real progress hunks ( 14:17 get abc?00e5d050). Observed cloning beagle.new (sub abc):

 14:17  new woof/test/STATIC.c          <- parent hunk (beagle format)
Enumerating objects: 2745, done.        <- git side-band band-2, RAW to stderr
Counting objects: 100% (2745/2745), done.
Compressing objects: 100% (907/907), done.
Total 2745 (delta 2096), reused 2415 (delta 1834), pack-reused 0 (from 0)
keeper: fetched 1 ref(s)                 <- keeper's own RAW stderr
 14:17  get abc?00e5d050                 <- back to parent hunk

Two raw-stderr sources bypass the hunk channel: git side-band band-2 is piped "to stderr live" by the ingest demux (keeper/KEEP.c:2381), and keeper prints fprintf(stderr, "keeper: fetched %u ref(s)\n", recorded) (keeper/WIRECLI.c:939).
Not a sub-only bug: the parent fetch emits the same raw text; the submodule recursion just makes it glaringly interleaved with the relayed parent hunks. Data correctness is unaffected — this is a progress-CHANNEL discipline gap.
Violates the producer/renderer split (Submodules/hunk model): progress should be hunk DATA (ts/verb/uri + text) rendered by the shared bro renderer, never a raw fprintf(stderr,...) from a producer.

Goals

Fetch progress (parent and submodule) flows as progress hunks through the one stream the parent relays, rendered uniformly — no raw git/keeper text on the terminal.

Constraints

Progress is hunk DATA, not presentation; the producer must not pick header style or write raw stderr (Submodules hunk split).
Capture git side-band band-2 and re-emit it as hunk text (or suppress under non-verbose) instead of live-to-stderr; band-3 (fatal) must still surface.
No pointer arithmetic / ABC style; keeper + sniff edits, not abc/.

WIP

Design decisions

—

TODOs

Route keeper fetch progress into the hunk stream: the band-2 demux (keeper/KEEP.c:2381) and keeper: fetched N ref(s) (keeper/WIRECLI.c:939) become hunk emissions, not fprintf(stderr,...).
Ensure the sub-recursion driver (sniff/SUBS.c SubMount) relays the child's hunk stream into the parent's, tagged with the sub URI so it reads like the parent's HH:MM get abc?… rows.
Repro-first (CLAUDE §17): be get a repo with a submodule over a fetching source; assert the report channel carries ONLY hunks — zero raw Enumerating/keeper: fetched lines.

Blockers and bummers

—