Add .NET 9 target, drop .NET 6

[MihaZupan]

Unlike in the past, there is now a reason to bump TFMs even when not explicitly using new APIs due to language changes that can target better overloads when recompiling (e.g. params span or OverloadResolutionPriority).
I don't know of any places where that would kick in for Markdig right now, but there are some neat new APIs in 9.0 that might be interesting to play with here.

I've also added back older TFMs to the test project so we're not completely without coverage (given we have a bit of conditionally-compiled code & polyfills). Skipped framework though cuz CI was being weird 🤷‍♂️

Some initial numbers without any Markdig changes on a large document (.NET 9 perf blog post ~570 KB):

Method	Runtime	Mean	Error
Parse	.NET 8.0	1,015.0 us	3.78 us
Parse	.NET 9.0	926.8 us	3.58 us
ParseAdvanced	.NET 8.0	2,892.6 us	17.57 us
ParseAdvanced	.NET 9.0	2,803.9 us	4.20 us
Render	.NET 8.0	459.2 us	5.76 us
Render	.NET 9.0	425.7 us	5.38 us
RenderAdvanced	.NET 8.0	585.2 us	10.13 us
RenderAdvanced	.NET 9.0	489.8 us	5.35 us

[MihaZupan]

Passing dotnet-version: '6.0 8.0 9.0' to the setup-dotnet action ends up building the latest LTS + the first item in the list.
With 6.0 8.0 that just happens to (currently) match the intended result.

[xoofx]

Thanks!

[snnz]

This is nice. But those whose projects are targeting .NET 6 will now receive a bonus with the updates, in a form of .NET Standard 2.1 DLL that misses some features of the .NET versions, like locals init switched off, and everything that depends on the "NET" condition.

[MihaZupan]

Such projects should likely focus on moving to a supported .NET runtime before updating their Markdig version then, especially if performance is a concern