host: keep a Session added while remove-agent awaits its archives - #2617

Written-by: Dev
Owner: Dev
Reviewer: Reviewer

Fixes #2616

remove-agent copied the DSH Agent's Session list, awaited one archiveSession per Session and
then wrote that pre-await copy back (record(agentId).sessions = kept), so a Session another
request added while it waited was dropped; when every archive succeeded the record was deleted
outright and the new Session was orphaned. The removal now applies its own outcome to the live
record: the kept branch filters out exactly the ids it archived (in rt.mutate), and the
all-archived branch keeps the Sessions the snapshot did not have and answers a retryable error
instead of deleting the record, so nothing is lost and a retry archives that Session.
archiveAgentSession, createAgentSession, adoptAgentSession and addAgent's rollback are
untouched; no compare-and-swap, no new endpoint, no UI change.

Ran at this tip: 645a9e9

$ cd core && DSH_HARNESS=/data/dsh/bot/workspace/d2607/repo/core/dev/harness TZ=UTC node --import ./test/harness-resolve.mjs --test test/agents.test.js
ok 9 - remove-agent drops only the Sessions it archived and keeps one added while it waited
ok 10 - remove-agent keeps the Agent and answers the retry error when a Session joins during the removal
# tests 16
# pass 16
# fail 0
exit 0

$ node scripts/ds-bot-patches.mjs check
ok   10 patches apply cleanly in order
ok   core/ equals ds-bot/ + patches/ds-bot/

$ node .github/scripts/changelog.mjs check
changelog: 18 fragment(s) in changelog.d/ are well formed

$ cd core && npx oxlint src test
Found 81 warnings and 0 errors.   (warnings pre-existing; lint tolerates them)

$ node core/scripts/build-client.mjs --check
exit 0

$ node .github/scripts/check-mojibake.mjs --check
mojibake: ok — 0 occurrences of U+FFFD in 1542 text file(s)

Not run: npm run verify (full suite; scoped to the target test), ds-bot-patches.mjs canonical
and arch-docs (both need git, absent on the dev box; core/ and patches/ were kept
byte-identical by check), smoke bench.

Red first: on the unmodified implementation the two new tests fail — ["s2"] instead of
["s2","s3"], and ok:true with the record deleted instead of the retryable error.

Checkpoint

{
"pr": 2617,
"claim": "Dev",
"issue": 2616,
"summary": "remove-agent no longer writes back the Session list it read before awaiting one archive per Session. The kept branch now filters the ids it actually archived out of the live record, and the all-archived branch keeps any Session added meanwhile and answers a retryable error instead of deleting the record, so a Session added during the removal is never dropped or orphaned. Two tests gate the archive hook and reproduce both interleavings; both were red before the fix.",
"files": [
"core/src/host/agents.js",
"core/test/agents.test.js",
"changelog.d/2616.fixed.md",
"docs/subsystems/core-host.md",
"patches/ds-bot/0001-core-fixes.patch"
],
"checks": { "ran": "node --import ./test/harness-resolve.mjs --test test/agents.test.js; node scripts/ds-bot-patches.mjs check; node .github/scripts/changelog.mjs check; npx oxlint src test; node core/scripts/build-client.mjs --check; node .github/scripts/check-mojibake.mjs --check", "result": "green", "at": "645a9e95cec3a136b4db4f24a280bcb7c0f6c43d" },
"next": "Reviewer: judge the tip for T2/T3. Nothing else pending."
}