Ported from the dev branch (192f6ab2, 4f013f44, dc0df1b5) into the package sdk/buildstep where
they were built; the package is not named by a design line (see below).
Builds: design 47 G §271 (a module claiming oci-artifact-registry but not its holder in the seat
list is refused), §24 (a step's end is published on its node's subject), §416 (the step ask carries
the kept results), §370 (the version).
Not decided: the package of SD2 and SD3. G-9 §420 names sdk/steprun only for the step-run code
(SD1); no line places the kept-artifact judgement or the step ask.
Claude-Session: https://claude.ai/code/session_01JG73KEq4zoDfM9WtYpcQgZ
A large node's request is 1.2 MB, over the bus's 1 MiB message limit, so the compose ask failed once keeping no longer did.
The checker packs the request (a JSON string "gzip+base64:..." in the part's `any` request, no contract change) and
buildstep.ComposeJob.RequestBytes reads either form. A kept request of another version is refused naming it, and listing
no longer reads pieces. An older mesh-build-go refuses a packed ask loudly (it holds no request object): the check fails
naming it, so the catalog change is delivered with this one as a group.
Quick fix (chosen as the recovery): pack the request, ~290 KB for the largest node. Thorough fix (milestone G): the request
named by digest and served on a seat verb, as RunInput is.
Claude-Session: https://claude.ai/code/session_01NcbgfFfQ4Z9UE5jK32PmE5
B0's mesh-sdk go/delivery run lines move to their registry homes: the
run job, the one-job rule, the Go check stream alone, the input's bound,
digest and key in sdk/buildstep; AskWhat's run case in sdk/buildrun; the
ask of the verb input and the check of a verdict's judged-by in
meshdeliverycheckerclient. B0's tests come in as rows of the existing
tables.
Claude-Session: https://claude.ai/code/session_01NcbgfFSnn24M3GcqttoU7i
jschoubben