fix(desktop): match WSAEADDRINUSE on Windows and repair the Go test build

[JasonBuildAI]

Summary

desktop/src's Go test package does not compile on develop (filedrop_test.go:91 redeclares err), and that compile error hides a real Windows-only defect from #1874: listenProbe matches busy ports with errors.Is(err, syscall.EADDRINUSE), but Windows reports a busy socket as Winsock errno 10048 (WSAEADDRINUSE) while Go's syscall.EADDRINUSE on Windows is the synthetic value 0x20000002, so the check can never be true. A busy preferred port (default 8088) therefore aborts firstFreePort instead of skipping ahead, and the desktop stops with a raw Winsock message instead of starting on 8089+ - the exact "second Octop instance" scenario #1874 set out to fix. Fixes #1905.

Target branch

• Base is develop (feature / fix - default)
• Base is main (release/* or hotfix/* only)

Type of change

• Bug fix
• New feature
• Breaking change
• Documentation
• Refactor / chore
• Release / hotfix

How it is fixed

• desktop/src/filedrop_test.go: drop the duplicate var err error in TestDropServerServesOnlyPublishedFiles, so go test ./... / go vet ./... build again (go build ./... was never affected because it ignores _test.go files).
• New addrInUse(err error) bool helper behind build tags: process_windows.go uses errors.Is(err, windows.WSAEADDRINUSE) (golang.org/x/sys/windows, already a direct dependency), process_unix.go keeps syscall.EADDRINUSE. listenProbe calls the helper.
• Errors that are not "address in use" (e.g. an unresolvable host) still propagate, so cannotBindError / noFreePortError handling in boot() is unchanged (TestChooseFreePortStopsWhenHostCannotBind still passes).
• New test TestChooseFreePortSkipsALiveListener binds a real listener and asserts chooseFreePort skips it - the user-visible path that regressed.

Test plan

Windows 11 x64, go1.25.9, worktree at 544ea757 (develop).

1. Before - pristine develop:

$ cd desktop/src && go test ./...
# octop.desktop [octop.desktop.test]
.\filedrop_test.go:91:6: err redeclared in this block
	.\filedrop_test.go:75:8: other declaration of err
FAIL	octop.desktop [build failed]
?   	octop.desktop/cmd/devserver	[no test files]

$ go vet ./...
vet.exe: .\filedrop_test.go:91:6: err redeclared in this block

2. Fixing only the duplicate declaration reveals the hidden Windows failure:

--- FAIL: TestListenProbeSeesOpenListener (0.00s)
    process_test.go:138: inUse = false, err = listen tcp 127.0.0.1:51360: bind: Only one usage of each socket address (protocol/network address/port) is normally permitted.
FAIL	octop.desktop	1.317s

Errno probe on the same host (scratch program, not committed):

busy bind err     = listen tcp 127.0.0.1:56430: bind: Only one usage of each socket address (protocol/network address/port) is normally permitted.
errors.Is(err, syscall.EADDRINUSE)          = false   // syscall.EADDRINUSE = 0x20000002
errors.Is(err, windows.WSAEADDRINUSE)       = true    // 0x2740 = 10048

3. Old process.go + the new tests (both fail, so the new test is a real regression guard):

--- FAIL: TestChooseFreePortSkipsALiveListener (0.02s)
    process_test.go:142: chooseFreePort(60857) = listen tcp 127.0.0.1:60857: bind: Only one usage ...
--- FAIL: TestListenProbeSeesOpenListener (0.00s)
    process_test.go:157: inUse = false, err = ...
FAIL	octop.desktop	0.825s

4. With the fix:

$ gofmt (LF-normalized copies of the five touched files) -> clean
$ go build ./... -> exit 0
$ go vet ./...   -> exit 0
$ go test ./...
ok  	octop.desktop	1.465s
?   	octop.desktop/cmd/devserver	[no test files]

$ go test -v -run 'TestListenProbeSeesOpenListener|TestChooseFreePortSkipsALiveListener|TestChooseFreePortStopsWhenHostCannotBind'
--- PASS: TestChooseFreePortStopsWhenHostCannotBind (0.06s)
--- PASS: TestChooseFreePortSkipsALiveListener (0.00s)
--- PASS: TestListenProbeSeesOpenListener (0.00s)
PASS
ok  	octop.desktop	0.208s

Python side (this change touches no Python code; make is unavailable on Windows, so the Makefile targets were run as their underlying commands):

• uv run ruff check src tests -> All checks passed!
• uv run ruff format --check src tests -> 1177 files already formatted
• uv run mypy src/octop -> Success: no issues found in 569 source files
• uv run pytest -n auto -m "not live" -q -> 28 failed, 4409 passed, 202 skipped in 26:42
• Baseline on the same host with the change stashed, same 10 files as the failures above: 27 failed, 67 passed, 1 skipped; the extra tests/unit/connectors/test_agently_auth.py::test_login_error_and_refresh_never_expose_cli_output failure in the full run passes standalone and under -n auto on the pristine tree (test-ordering/resource flake). The 27 environment failures are pre-existing (tests/unit/db/* legacy-migration fixtures and tests/unit/test_vendor_wheels_argv.py, the latter tracked by test(desktop): apply posix_only to the vendor-wheels argv cases #1812).

Checklist

• Ship-bar commands run locally (see Test plan; make is unavailable in this Windows clone, so the Makefile targets were run as their underlying commands). ruff check / ruff format --check / mypy pass; the pytest -m "not live" run reports 28 failures, all pre-existing on this host (27 reproduced with this change stashed, plus one xdist flake), and this PR touches no Python file.
• Added/updated tests
• Updated CHANGELOG.md
• README / docs updated (not needed)

Note for reviewers, not part of this PR: ci.yml has no Go test job and octop-desktop.yml only packages the app, which is why a non-compiling test package sat unnoticed since #1874. Adding a go vet ./... + go test ./... job for desktop/src would catch this class of regression; happy to follow up separately if wanted.

AI disclosure: this change was drafted with AI assistance at my request, under my direction and review.