Template selection regression

[jmooring]

This is a regression from 0.147.9 to 0.148.0.

Git bisect says cfc8d31 is the first bad commit. I made this change to address #13826, and it is obviously missing some test cases.

project/
├── content/
│   └── p1.md
├── layouts/
│   ├── p1/
│   │   └── page.html  <-- v0.147.9 uses this (correct)
│   └── page.html      <-- v0.148.0 uses this (incorrect)
└── hugo.toml

func TestIssue13868(t *testing.T) {
	t.Parallel()

	files := `
-- hugo.toml --
disableKinds = ['home','rss','section','sitemap','taxonomy','term']
-- content/p1.md --
---
title: p1
---
-- layouts/page.html --
layouts/page.html
-- layouts/p1/page.html --
layouts/p1/page.html
`

	b := hugolib.Test(t, files)

	b.AssertFileContent("public/p1/index.html", "layouts/p1/page.html")
}

Reference: https://discourse.gohugo.io/t/custom-layout-not-working-for-contact-md/55405

[jmooring]

We may need to revert cfc8d31. While putting together a more comprehensive integration test, my initial take is that we can't have it both ways. I suspect bep ran into this when implementing the new template system in v0.146.0.

[bep]

We may need to revert cfc8d31.

Lets do that for now, as this certainly is a worse issue than that one. I'll also try to get a patch release out the door with this and #13864 relatively fast.