Pre-allocate len of FixedVector during Decode

[paulhauner]

Use Vec::with_capacity when decoding a FixedVector, based on num_items.

num_items has been checked against N::to_usize() (where N is the fixed length of the vector). This protects us against memory explosions, as is common with pre-allocations.

When flame-graphing SSZ block decoding we can see that growing the logs_bloom FixedVector during ExecutionPayload decoding takes some time.

Notably, this refactor didn't make a significant improvement in block decoding times. However I still think it's worth including because it is kinder to the machine.

Benchmarking

Here's the flamegraph from before this change was implemented. It requires some searching to find the SSZ decode function, but it's there. Screenshots below.

Entire Block Decoding

Zoomed into FixedVector::from_ssz_bytes

[michaelsproul]

Looks great!