Reduce TrieNode allocations

[Scooletz]

This PR removes 10% managed allocations from the node. It does it by moving StorageRoot from TrieNode as an item of an array and getting rid of RlpStream from TrieNode and delegating its functionality to Rlp.ValueDecoderContext

Tests

Benchmarks

Reused PatriciaTreeBenchmarks that provides a coarse grained measurement of various operations.

BenchmarkDotNet=v0.13.5, OS=Windows 11 (10.0.22621.1413/22H2/2022Update/SunValley2)
12th Gen Intel Core i9-12900HK, 1 CPU, 20 logical and 14 physical cores
.NET SDK=7.0.100
  [Host]     : .NET 7.0.0 (7.0.22.51805), X64 RyuJIT AVX2
  DefaultJob : .NET 7.0.0 (7.0.22.51805), X64 RyuJIT AVX2

Method	Mean	Error	StdDev	Median	Gen0	Gen1	Allocated
master	245.4 us	8.13 us	23.84 us	254.4 us	12.4512	3.1738	155.43 KB
this PR	181.5 us	9.02 us	26.59 us	193.7 us	10.2539	2.6855	126.38 KB

Running node

Version	Memory allocation - min	Memory allocation - max	Memory allocation - avg	New Payload Execution Time - Mean	New Payload Execution Time - Stdev
master	4.11 MB/s	21.8 MB/s	5.46 MB/s	891	496
this PR	3.90 MB/s	12.5 MB/s	4.80 MB/s	779	408

**Image** comparing of all the metrics

Changes

• StorageRoot is now stored as _data[2] only in leafs
• RlpStreamReader used instead of the object of RlpStream

Types of changes

What types of changes does your code introduce?

• Bugfix (a non-breaking change that fixes an issue)
• New feature (a non-breaking change that adds functionality)
• Breaking change (a change that causes existing functionality not to work as expected)
• Optimization
• Refactoring
• Documentation update
• Build-related changes
• Other: Description

Testing

Requires testing

• Yes
• No

If yes, did you write tests?

• Yes
• No

Remarks

Another approach to the TrieNode optimization.

[Scooletz]

Maybe I can change the implementation to use Rlp.ValueDecoderContext?

[asdacap]

If you want to micro op, I guess you can set this to a byte. Or it does not matter and overhead of casting it is higher.

[Scooletz]

I think I'll leave as is. The stack is the stack one int or byte wont' make that much change. Also, I'll default to Rlp.ValueContext for now.

[Scooletz]

Doing a full sync on teku mainnet. Once it's synced and working will confirm and merge if the results are consistent.

[Scooletz]

The teku mainnet has been synced with this and now is operational. Waiting for a few hours of running and will be merging if nothing contradicts the findings above.

[Scooletz]

I confirm that with the usage of the Rlp context reading the version behaves as it was with the copied RlpStreamReader, reducing the measured alloc ratio by at least 10%. Merging.