docs: clear shipped plans and closed work out of the documents

Owner, 2026-10-18: "Надо почистить документы от устаревшего и сделанного."

- Deleted, shipped: plans/delivery-log-storage, from-header-binding,
  inbound-antispam-panel, inbound-upstream-result, panel-dkim-signing,
  panel-notifications, and the own-mta-builds analysis (its three rows
  shipped as 2.6.1, 2.7.0, 2.8.0). What outlived them moved first: when a
  filter-list counter is worth a migration, and the notifications' "when"
  table, into architecture.md; the sender check's own-process trigger into
  security.md. Code comments that cited them point at architecture.md or
  security.md; CHANGELOG keeps their names, unlinked.
- Kept, marked closed: audit-2026-10-12, assumptions-2026-10-14 and
  security-audit-2026-10-17 - the code cites their finding ids in 50-84
  files. Their file:line refs are of the tree each audited.
- roadmap.md: open work only - stage H of the security audit (the owner's),
  go-milter-upstream (the owner posts), and queue-table, which the
  redesign left undecided and only its own shipped section still named;
  Dropped and Ideas as they were. 2.x versioning instead of 1.x.
- development.md: the 47-row planned-version-cuts table (history, in the
  CHANGELOG) gone; a stage's last step now removes its roadmap item and
  plan. Docs map lists the closed audits.
- panel-redesign.md: the done checklist out; Room for roadmap items says
  which drawn screens are unbuilt and why.
- Stale facts fixed: schema-migrations.md and architecture.md at schema 14
  (0013's users columns and user_links, 0014's secure TLS mode);
  security.md's accepted risk "profile saved without the password" - the
  e-mail asks for it since 2.14.0.

Tests: go build, go vet; go test of cmd and the touched internal packages
pass except logtail's three Windows-bound rotation tests (rename of an open
file; CI is Linux). Only comments changed in Go.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>

build/Dockerfile

@@ -57,7 +57,7 @@ ENV DEBIAN_FRONTEND=noninteractive
 # test say whether it does.
 # No DKIM package is installed: the panel signs itself (internal/milter,
 # sign.go) and makes the keys in Go (internal/domain/dkim.go); OpenDKIM left
-# the image in 2.8.0 (docs/plans/panel-dkim-signing.md).
+# the image in 2.8.0 (docs/architecture.md § DKIM signing).
 # systemd-standalone-sysusers is named so that cron-daemon-common, which
 # logrotate brings (logrotate 3.22.0-1, debian/control: "Depends: cron |
 # anacron | cron-daemon | systemd-sysv"), creates its group with it rather

build/entrypoint.sh

@@ -111,7 +111,7 @@ find /data -mindepth 1 -maxdepth 1 ! -name log ! -name postfix ! -name sasl ! -n
 # serves (cmd/panel/signer.go). Until 2.8.0 the keys were under
 # /data/opendkim, shared with OpenDKIM; 2.8.x copied them to /data/dkim at its
 # first start and kept the old tree for a rollback to 2.7.x until the next
-# release, this one (docs/plans/panel-dkim-signing.md, Q2). It is removed
+# release, this one (CHANGELOG [2.9.0]). It is removed
 # once /data/dkim holds a key. A /data that never ran 2.8.x - keys under the
 # old tree only - stops the start instead: the panel would make the server's
 # own name a new key, and its published record would no longer match; 2.8.x

build/postfix-config.sh

@@ -426,7 +426,7 @@ postconf -e \
 # deferred (docs/security.md).
 #
 # The panel's DKIM signer follows, strict too: unreachable, it defers rather
-# than send unsigned (docs/plans/panel-dkim-signing.md). It signs with the
+# than send unsigned (docs/architecture.md § DKIM signing). It signs with the
 # key of the login's domain. The journal-milter is monitoring only and is
 # fail-open (default_action=accept): its failure must never block the relay.
 # Per-milter settings use Postfix 3.0+ brace syntax.
@@ -756,7 +756,7 @@ if [ "${INBOUND_RELAY_ENABLE}" = "true" ] || [ "${DMARC_REPORTS_ENABLE}" = "true
 	# only here (above), so no client is spared. The submission ports take
 	# the default, normalize as well since this Postfix: an application whose
 	# SMTP library ends DATA with a bare <LF>.<LF> is no longer read to its
-	# end there (own-mta-builds, § 5: their smuggled transaction would still
+	# end there (docs/security.md § Mail path: their smuggled transaction would still
 	# be the authenticated session's, so they were left to the default).
 	postconf -P \
 		"smtp/inet/smtpd_forbid_bare_newline=normalize" \