Clean up render snippets

[LHolten]

Hey, here are some local changes I made that might be useful.

In the first commit I tried to make the render_snippets function more readable and fixed what I think was a bug.
(Looks like it would not merge more than two contexts if the last context didn't overlap the first)

In the second commit I restricted finding the primary label to those in the current context.
This makes it so that the header above each snippet points to code in the snippet.
(It is especially useful for SourceCode implementations that wrap multiple files)

Example showing different headers for different snippets:

[Benjamin-L]

I'm not confident that the x_conts.line(), x_conts.line_count() -> x_line, x_line_count change is actually more readable personally, but don't feel strongly about it. It would probably also be good to add a test for the combined-context bug in tests/graphical.rs. The rest of this looks great to me.

[LHolten]

I'm not confident that the x_conts.line(), x_conts.line_count() -> x_line, x_line_count change is actually more readable personally, but don't feel strongly about it.

I thought this change was necessary because I removed the vector of SpanContents. Turns out it was not necessary so I reverted the change and it is now even more readable!

[LHolten]

Also added a test for non-adjacent contexts, which shows that each context gets a header that points to the code in the context :D

miette/tests/graphical.rs

Lines 1791 to 1802 in e16326a

	╭─[bad_file.rs:1:1]
	1 │ source
	· ───┬──
	· ╰── this bit here
	2 │
	╰────
	╭─[bad_file.rs:5:3]
	4 │
	5 │ text here
	· ──┬─
	· ╰── also this bit
	╰────

[Benjamin-L]

New version looks really good to me :)

[zkat]

Thanks for this!