fix: wrong syntax highlighting for embedded languages

[jfcherng]

Fixes https://github.com/TheSecEng/LSP-copilot/issues/33

---

For example, syntax highlighting for completing JS codes inside Markdown fenced code block.

The main idea is to get the deepest/last source | text | embedding scope and find the languageId for it.

For the example above, the scope at the cursor is text.html.markdown markup.raw.code-fence.javascript.markdown-gfm source.js. We should take source.js for resolving languageId rather than text.html.markdown.

[jfcherng]

Unfortunately, this won't work for PHP.

Python 3.3>>> from LSP.plugin.core.types import basescope2languageid
Python 3.3>>> basescope2languageid('source.php.embedded.html')
'html'

[jfcherng]

I notice that

• PHP uses source.php.embedded.html.
• Markdown uses source.json.embedded.markdown. (https://github.com/sublimehq/Packages/blob/17d27049bcc41cb15561476743462c46968fc82b/Markdown/Markdown.sublime-syntax#L317)

So I assume that we have a scoping convention for embedded things: EMBEDDED_LANG.embedded.PARENT_LANG. Hence we can extract EMBEDDED_LANG, which we are interested in.

[jfcherng]

Looks good again.