docs: add a troubleshooting section and document how a converge runs

[tas50]

The README documents every option, but it never says what to do when a converge fails — which is when people actually open it. This fills that in.

What changed

A "How it works" section. Five numbered steps: set the LCM, prepare the instance and install gallery modules, stage files into the sandbox, compile the MOF, apply it. Two reasons for it — the option tables below now have somewhere to hang, and a converge that dies partway through can be traced to a named step.

A "Troubleshooting" section. Written from the errors this provisioner can actually produce, not generic advice. It covers the two throw messages we generate on the instance (Failed to find <path>, Failed to create a configuration command <name>), the local ENOENT before the instance is even touched, and the resource that never reached the PSModulePath.

The three I most wanted written down are the ones that fail silently:

• modules_from_gallery is ignored unless dsc_local_configuration_manager_version is exactly wmf5. On wmf4 nothing installs and nothing is logged.
• Module style is only detected when the kitchen root contains <directory name>.psd1 — a manifest named after the directory you cloned into. Clone xWebAdministration into a folder called webadmin and it silently falls back to repository style.
• gallery_name on its own only works if that package source is already registered on the instance.

A Type column on every option table, plus the resolved behaviour that was previously only discoverable by reading the source:

• which gallery name you actually get with none, one, or both of gallery_name and gallery_uri (PSGallery, testing, or yours)
• that an unrecognised LCM version falls back to the base WMF 4 settings without warning
• that the reboot defaults (retry_on_exit_code: [35], max_retries: 3) are only applied if you left Test Kitchen's own defaults alone
• that a missing modules_path is not an error

A Contents list, matching kitchen-docker.

Also picks up the one-line gemspec spacing fix from #115 so lint is green here too — cookstyle 9 flags it and it currently fails on every PR. Whichever lands first, the other becomes a no-op.

Verification

$ npx markdownlint-cli2 --config .markdownlint.yaml README.md
Summary: 0 issues in 0 files

$ bundle exec cookstyle --chefstyle
9 files inspected, no offenses detected

$ bundle exec rake test
68 examples, 0 failures