doyle -> todlando: HOLD W4 for one thin lane. Do not branch feat/339-update-sources yet. I found a defect in
what I just landed, and it is mine, not yours.

THE DEFECT. At f6a55013, cli.rs:1461-1466 the flag is declared `#[arg(long = "via-subnet")]` with NO alias of
any kind -- I grepped the whole file for `alias = "vs"` and `short_alias`: zero hits. But the doc comment on
that same arg says "(also spelled `-vs`)", and because the help string is the source of the generated docs,
the PUBLIC reference.md now says it too, at docs-site/src/cli/reference.md around line 167. So as of ten
minutes ago spt ships public documentation for a flag spelling that does not exist. A user who reads the docs
and types it gets an error.

HOW IT GOT THROUGH ME. I audited that exact help string twice today -- once hunting tracker codes, once in
your fix diff -- and read straight past it both times, because I was checking whether the docs MATCHED the
source and never asked whether the source was TRUE. xtask check cannot catch this either: its docs-drift rule
compares generated text to source text, and here they agree perfectly and are both wrong. That is the whole
mechanism, and it is worth more than the fix.

IT ALSO MEANS #264 IS NOT DELIVERED. #264 asks for `spt adapter add <adapter-name> --via-subnet|-vs`. The long
flag landed; the alias half did not. So #264 does NOT go to acceptance today -- only #322 does.

THE FIX, ruled. `-vs` is not buildable as written: clap shorts are single characters, so a two-character `-vs`
short cannot exist. The achievable spelling is `--vs` as a long alias, and it must be VISIBLE, because this
file already records at cli.rs:118-123 that a plain `alias` is hidden and does NOT render into reference.md --
which would leave the docs silent about a flag that works, the same defect mirrored.

  #[arg(long = "via-subnet", visible_alias = "vs")]

and correct the sentence to say `--vs`, not `-vs`. Then `xtask gen` with CARGO_TARGET_DIR unset, read the
diff, and confirm reference.md renders the alias. That makes the help TRUE and delivers #264's ask as closely
as clap allows.

LANE SHAPE. A thin lane of its own off f6a55013, pushed and landed before W4 -- not a rider in W4's PR, which
is #339's lane and must not carry #264's surface. It is a two-line change plus a regen, so it is minutes, and
landing it first means you branch W4 off the newest main and STILL rebase zero times. Tag the REQ evidence for
whichever REQ carries the --via-subnet surface and run treqs before you push.

Send me the sha when CI is green on it and I will land it, then release you onto W4.