perf: use ValueListBuilder<T> for constant buffers

[TimothyMakkison]

Follow on from #1502, using ValueListBuilder to replace known List<Doc> sizes.

• Most of these changes aren't measurable in the benchmarks, many are insignificant, although I suspect this is partly due to the limited code found in the benchmark file.
• This PR will increase the performance benefits of perf: add new Concat types for children lengths #1524

@belav what do you think the best way of replacing a list that uses a collection initializer is?

I can either add the items directly into the scratch buffer and manually update Length

var docs = new ValueListBuilder<Doc>
([
    Modifiers.Print(node.Modifiers, context),
    Token.Print(node.DelegateKeyword, context),
    null,null
]);
docs.Length = 2;

Or create the buffer and then append the items via params.

var docs = new ValueListBuilder<Doc>([null, null, null, null]);
docs.Append(
    Modifiers.Print(node.Modifiers, context),
    Token.Print(node.DelegateKeyword, context)
);

Or Append each item

var docs = new ValueListBuilder<Doc>([null, null, null, null]);
docs.Append(Modifiers.Print(node.Modifiers, context));
docs.Append(Token.Print(node.DelegateKeyword, context));

[TimothyMakkison]

• Opted to append each item to keep similar formatting
• Double checked the scratch buffer sizes for each method
• Added Doc.Concat(ref ValueListBuilder<Doc>) for the token and separated syntax list methods. It might be worth adding AggressiveInlining to it 🤔

Nice little performance gain and saves 1.5 MB on the complex code benchmark, it used to be larger before #1535

I'll look into doing this for the variable size buffers later.

Before

Method	Mean	Error	StdDev	Gen0	Gen1	Gen2	Allocated
Default_CodeFormatter_Tests	102.3 ms	2.04 ms	1.91 ms	3333.3333	2333.3333	666.6667	31.13 MB
Default_CodeFormatter_Complex	173.5 ms	2.52 ms	3.99 ms	5000.0000	2000.0000	-	54.57 MB

After

Method	Mean	Error	StdDev	Median	Gen0	Gen1	Gen2	Allocated
Default_CodeFormatter_Tests	95.02 ms	1.878 ms	3.793 ms	93.56 ms	3000.0000	1666.6667	333.3333	30.81 MB
Default_CodeFormatter_Complex	168.67 ms	1.949 ms	2.795 ms	168.72 ms	5000.0000	2000.0000	-	53.08 MB

When combined with #1524 (timing is inaccurate)

Method	Mean	Error	StdDev	Median	Gen0	Gen1	Allocated
Default_CodeFormatter_Tests	127.0 ms	2.33 ms	3.49 ms	127.3 ms	2000.0000	1000.0000	29.21 MB
Default_CodeFormatter_Complex	264.6 ms	6.40 ms	18.88 ms	271.5 ms	4000.0000	2000.0000	49.97 MB

[belav]

Opted to append each item to keep similar formatting

That is the option I would have gone for. I think it keeps it cleaner. I also like consolidating the logic for dealing with Concat with 0, 1 or more items. Nice work!