test(cmd/gc): use fake session provider in Phase0 spec tests (ga-kkn)

[scarson]

Summary

Fixes the macOS test flake tracked as bead ga-kkn: five TestPhase0*
cases in cmd/gc/session_model_phase0_spec_test.go and one in
cmd/gc/session_model_phase0_factory_namespace_spec_test.go set
t.Setenv("GC_SESSION", "phase0"). The cmd/gc provider context
(loadSessionProviderContext in cmd/gc/providers.go) treats
GC_SESSION as the provider-name override. "phase0" is not a known
provider, so the runtime falls through to the default tmux provider.

When the developer's host already has a real tmux session named
mayor (or any matching named-session target), the provider's
session-create path returns runtime.ErrSessionExists and
TestPhase0CanonicalMetadata_NamedMaterializationWritesNamedOriginWithoutLegacyManualFlag
fails with:

session name already exists: "mayor" already active in runtime

Switch the sentinel to "fake" so these tests use runtime.NewFake()
and never touch the host's tmux server, matching the convention
already established across cmd_session_*_test.go, cmd_mail_test.go,
cmd_nudge_test.go, etc. — every other test in the package that needs
to control the session-provider env var sets it to "fake".

Trail

• test: skip 127.0.0.0/8 alias bind on darwin (1 of 2 pre-existing macOS flakes) #2063 (gascity/flake-inv → sa-9sa3fn investigation) — fixed
  TestResolveDoltConnectionTargetManagedCity_EnvOverride and
  identified two real underlying flakes that PRs fix(bd): reap stale .beads/issues.jsonl on managed scopes #2060/fix(cmd/gc): prefer rig binding over findCity legacy fallback for GC_DIR #2062 had
  misattributed.
• bead ga-kkn — this one.
• bead ga-un5 (gascity/perf-inv) — the sibling cmd/gc perf flake.

Fix choice (vs alternatives)

The bead listed candidate fixes; I picked the smallest-change option:
swap the sentinel value. Alternatives considered:

• (picked) Use the existing \"fake\" sentinel. 5-line diff, zero new
  helpers, aligns with prevailing convention. Root cause is that
  \"phase0\" was a typo'd sentinel that silently fell through to tmux.
• Inject a provider through the function signature
  (resolveSessionIDMaterializingNamed, friends) — much larger
  refactor; the lookup path already supports \"fake\" via env var.
• Build a separate test-only provider context — unnecessary; the
  \"fake\" path through newSessionProvider already exists for
  exactly this purpose.

Repro

Before, with a host tmux session named `mayor`:

tmux new-session -d -s mayor
go test -count=1 -run TestPhase0CanonicalMetadata_NamedMaterializationWritesNamedOriginWithoutLegacyManualFlag ./cmd/gc/

=> FAIL: session name already exists: \"mayor\" already active in runtime

After:

go test -count=1 -run TestPhase0CanonicalMetadata_NamedMaterializationWritesNamedOriginWithoutLegacyManualFlag ./cmd/gc/

=> ok

Full go test -count=1 ./cmd/gc/ passes locally with this change.

Testing

• go test -count=1 ./cmd/gc/ — green
• Bead's repro recipe — no longer fails
• go vet ./cmd/gc/... — clean
• go build ./... — clean

Pre-commit hook note

The local .githooks/pre-commit make test gate trips on three
pre-existing failures on origin/main (1c5b6073) that are unrelated
to this change:

1. internal/beads/contract.TestResolveDoltConnectionTargetManagedCity_EnvOverride
   — the macOS 127.0.0.2 bind issue that PR test: skip 127.0.0.0/8 alias bind on darwin (1 of 2 pre-existing macOS flakes) #2063 fixes (still open).
2. cmd/gc.TestCityRuntimeReloadDrainShortCircuitsOnTickContextCancel
   — flaky under parallel make test, passes standalone. Forwarded
   to gascity/perf-inv as a sibling to ga-un5.
3. internal/runtime/acp.TestStartLongSocketPathUsesShortSocketName
   — flaky under parallel make test, passes standalone. Filed as
   bead ga-urt.

All three reproduce on a stashed-clean origin/main checkout in this
worktree; this PR touches only *_test.go files so it cannot affect
them. Verified via git stash && go test -count=1 -run ... && git stash pop.

Checklist

• Linked the bead (ga-kkn) and related PR (test: skip 127.0.0.0/8 alias bind on darwin (1 of 2 pre-existing macOS flakes) #2063) in the body
• Test-only change; no production behavior changes
• No docs to update (test sentinel rename)
• No breaking changes

🤖 Generated with Claude Code

[julianknutsen]

Maintainer Adoption Review

Thanks for the contribution, @scarson! This PR switches the affected Phase0 session-model tests from the phase0 GC_SESSION sentinel to the registered fake provider, which keeps those specs on the in-memory runtime instead of accidentally touching a developer's host tmux sessions.

This PR was reviewed and adopted with maintainer fixes pushed directly to the PR branch.

Original PR Review

Decision: approve.

Specific gaps fixed:

• The Phase0 specs were using GC_SESSION=phase0, which is not a registered provider name and could fall through to the default tmux provider. The final patch uses GC_SESSION=fake in the five affected setup lines so the tests stay isolated from host tmux state.
• Review checked the scope of the sentinel update and confirmed it is limited to cmd/gc/session_model_phase0_factory_namespace_spec_test.go and cmd/gc/session_model_phase0_spec_test.go, with no production behavior change.
• Focused validation covered the changed surface: the broader go test -run 'TestPhase0' ./cmd/gc suite passed locally on the refreshed head, and the pushed head also has passing visible GitHub CI.

Review findings addressed:

• No new review findings required repair. The review validated the tmux collision mechanism, the fake-provider routing behavior, and the narrow two-file test-only diff.

Remaining non-gating notes:

• No remaining gating issues were found.

Maintainer Changes

Maintainer-side update pushed directly to the PR branch: rebased the reviewed commit test(cmd/gc): use fake session provider in Phase0 spec tests (ga-kkn) onto current main at e6518555, preserving the contributor author and stable patch-id. Short diff stat: 2 files changed, 5 insertions(+), 5 deletions(-).

Final Review Status

Ready for the merge queue: final head 0d44f09 has passing required GitHub checks, and the merge-ready metadata is in place. I am marking this PR status/merge-ready so the merge queue can pick it up.

CI: https://github.com/gastownhall/gascity/actions/runs/25893356324

Review Iterations

1 review pass performed. The review/fix loop approved the test-only fake-provider update; finalize replayed the patch onto current main, reran focused Phase0 validation, pushed the reviewed head to the PR branch, and observed passing GitHub CI.

---

Adopted via /adopt-pr workflow. Original contributor commits preserved.