docs: Resolve the committed merge conflict in the doctor section

What this MR does and why

.claude/skills/glci/SKILL.md had unresolved git conflict markers committed into the ### glci doctor section in 5fda6de, and they are on main today.

This is not confined to the repository. glci skill install fetches this file from the repository at runtime (fetchSkillForBuild in cmd/glci/skill.go) at the build commit, falling back to main — so every install writes the markers into the user's agent skill directory (.claude/skills/glci/, .gitlab/duo/skills/glci/, .agents/skills/glci/, .opencode/skills/glci/) and the glci doctor guidance is unusable for the agent reading it.

The resolution takes the newer Podman-aware side as the base and carries forward the one fact only the HEAD side had: the Daemon endpoint row and its drift warning, which cmd/glci/doctor.go:311-340 confirms is still live. The result was checked against cmd/glci/doctor.go — check order (runDoctor, :136-153), the ✓/✗/~ symbols and the unsymboled → continuation (:97-119), the exit-1 conditions, colour gating (:83-95), and the Podman socket cases (:226-236) — and against site/content/cli-reference/commands.md:441-488, which was never corrupted and describes the same behaviour.

Docs only; no code changes.

Steps to reproduce

1. On main: grep -n '<<<<<<<\|>>>>>>>' .claude/skills/glci/SKILL.md — two marker lines are reported.
2. From another project directory: glci skill install --agent claude-code --scope project.
3. grep -n '<<<<<<<' .claude/skills/glci/SKILL.md in that project — the installed skill carries the same markers.

After this MR, step 1 reports nothing and the ### glci doctor section reads as one coherent block.

Verified locally: go build ./..., go vet ./..., make test (all packages), and the unit CI job run with glci run unit. A repository-wide scan finds no remaining conflict markers.

Relevant issues and other links

Closes #134 (closed)