fix: peer template review follow-ups

[floreabogdan]

Follow-ups from the review of #21 (peer templates). Each finding was checked against main first. The name clash was reproduced with real bird -p. Each fix has a test that was watched failing. One commit per fix:

Fix	Commit
Template names could clash with any BIRD symbol. template bgp TRANSIT beside protocol bgp TRANSIT, a set, a community or a generated name failed every apply with "Symbol already defined". store.SymbolUses finds every owner of a name; template saves refuse a taken name, and peer saves refuse a template's.	af861d8
A community only a template referenced could be deleted, leaving every later-attached peer with an undefined symbol. Templates now count as users.	0e3f540
A drained peer that became iBGP through a template kept its drain, so the iBGP filters set local-pref 0 and GRACEFUL_SHUTDOWN. updatePeerShape clears it for iBGP, like validateShape.	51baf44
Folded lint findings lost the peer names and counted findings, not peers. They now read IX_PEERS: rs3_v4, rs7_v4 and fold only across 2+ distinct peers.	cec6003
The bulk bar defaulted to "Attach to ", and detach posted an empty value. There's now a required, disabled placeholder and an explicit detach; an empty choice is refused.	40370ad
Link and bulk attach read outside their transaction, and the bulk bar committed one transaction per peer and audited after the loop. store.AttachPeers does the whole selection in one transaction, reading inside it; link and detach are its one-peer cases.	edca30b
iBGP template previews used AS64496, so every one showed a false "marked iBGP" danger. The sample name also exceeded 63 chars for long template names.	bc78ff0
A failed rename re-rendered posting to the typed name, so the resubmit hit "not found". Fixed for templates and for the peer form, which had the same bug.	4054ed2
The policies list ignored template chains, showing "nothing" for a policy delete refuses.	93730b9
Every failed template delete said "detach those peers first". It's now a typed TemplateInUseError, and only that gets the hint.	904b757

Not changed, deliberately:

• Peers already named templates/attach get no migration. That's vanishingly rare, and renaming would restart live BGP sessions.
• Linked peers aren't rendered from the template row (review item 14). That's a design change to "the template is not consulted at render time".
• Preview query counts (item 15) are left alone.
• The peer form's own read-then-save of a linked peer, and import-from-BIRD's create-then-chain, keep their existing two-step shape.

Verification. Actions is billing-locked, so this was run locally with the ci.yml steps (go1.26.8, GOTOOLCHAIN=local): gofmt clean, vet (including -tags integration) ok, golangci-lint 0 issues, go test -race ./... ok, govulncheck with nothing reachable, the build ok, and both BIRD 2.14 integration tests passing.

🤖 Generated with Claude Code

[floreabogdan]

Review follow-up (pushed):

• Namespace check, at the right altitude. The renderer now refuses any model that declares one BIRD symbol twice, naming both declarations, plus BIRD's own master4/master6. That covers every creation path, including import, restores and data that already clashes. Prefix sets, AS sets, communities, RPKI servers, BMP stations and import-from-BIRD also refuse a template's name at save (6e6b97e, bfcd263).
• Derived names clash only while their source exists. exp_transit is free unless a policy transit exists. Community validation uses the shared built-in list instead of its own shorter copy (147c897).
• AttachPeers writes first. It touches the template row to take SQLite's write lock up front instead of read-then-upgrade, which risks SQLITE_BUSY_SNAPSHOT. It also reads set IDs through its transaction: policySetIDs takes the querier, and a pool-of-1 test hung without it. LinkPeerToTemplate(peer, 0) is ErrNotFound, not a detach. The unused DetachPeer is removed. A mid-request deletion in the bulk bar says "nothing was changed" instead of returning a 500 (b2835d5).
• The community delete guard now refuses when its in-use check fails (8fef211).
• All the form's name links use the stored name after a failed rename (da9cebc).
• The policies list counts template use with one GROUP BY (96418a6).

Left as is:

• Two saves of the same new name racing between check and insert. The render check now catches that before any apply.
• The duplicate settings read in the live preview.

Re-verified locally: gofmt, vet (including -tags integration), golangci-lint, go test -race ./..., govulncheck, the build, and BIRD 2.14 integration all pass.