Delete the installer host wiring neither installer runs (#691)

[MongLong0214]

Closes #691.

install.ps1   1171 -> 820 lines   (-351, everything after an unconditional exit)
install.sh    1211 ->  717 lines   (-494, has_/wire_ pairs nothing calls, and their helpers)

Why this is not the risky deletion the issue described

install.ps1:817-820 delegates and exits:

$hostSummary = & $dest installer-hosts --wrapper $dest --data-root $dataRoot --home $env:USERPROFILE --json
$hostExit = $LASTEXITCODE
$hostSummary | Write-Output
exit $hostExit

Unindented, unconditional, since 0d60c75 (#624). Everything after it is unreachable. install.sh:1035 delegates the same way and its wire_* functions are never called.

This issue's step 2 was gated on a Windows machine because the verification was framed as an install with those blocks removed must leave a machine wired the same way — which assumed the blocks take part in wiring. They cannot. No run can distinguish their presence from their absence, so the gate does not apply to deleting them. What still needs Windows is whether the enumeration wires a real machine correctly, and that ships either way.

What it was costing

#689 happened because install.sh read as the list of supported hosts. This issue's body records two more readers reaching the same conclusion the same day. Today I read install.ps1:881 as live behaviour and filed #712 on it, in response to an outside report, before checking control flow — three readers in three days, once past the repository boundary.

Guards that were asserting against the dead copies

Four moved to the code that runs:

guard	was	now
#689 host coverage	compared install.sh pairs to the CLI	calls inspectAndApplyHosts and requires every known host in notDetected
T-1121 parity	asserted ps1 mirrors sh on wiring	parity for what both still implement; wiring asserted against the enumeration
#660 plugin upgrade	awk-extracted wire_claude_code from install.sh	runs the enumeration with a fake claude on PATH
T-1123 config table	required both installers to spell each path	requires the enumeration to

The #689 rewrite is the one that mattered. With the shell gone, a host list read out of the module under test would shrink with it and report nothing — so the list is written down in the test and the check is behavioural. A branch that returns early without pushing is missing code: no text search can see it.

Negative control: removing the windsurf row from the enumeration fails the guard with windsurf appears in neither hosts nor notDetected — it was dropped, not skipped.

Verified

• 142 cases pass across the seven installer suites; npx tsc --noEmit clean; sh -n install.sh clean
• npm run build:canonical then artifact:verify → de9222c7…, dist unchanged — no src in this change
• test/init.test.ts fails 6 cases on clean main here and on this branch alike; a local environment difference, not this change. CI is the authority and is green on main

---

Scope of this closing. #691 held two claims: that deleting these blocks cannot change an install (static — this PR), and that installer-hosts wires a real Windows machine (never verified). Only the first is closed here. The second moved to #714 and remains open, and is sharper after this change: with the shell copy gone, the unverified path is the only path.

[github-actions]

CommitLore — record lint

Trailers: clean — 1 commit in origin/main..7c763d979ab6e24a6736a71c4d9f9204e2177aab
Active constraints: not read — commitlore: git log --follow accepts exactly one pathspec, so renames are not followed for 7 paths; query one path at a time to follow its rename chain (7 changed paths)

_{Trailer violations fail this check. Active constraints are informational — they are what the repository already decided, not a verdict on this PR.}