Speed up reads by adding support for lazy decoding of registers

[fxamacker]

Issue To Be Solved

Decoding should be optimized to improve speed. Currently, Atree encodes in parallel (multiple goroutines) but decodes sequentially.

Suggested Solution

Speed up reads by adding support for lazy decoding of registers.

NOTE: It might be possible to also speedup writes but that will most likely be opened as separate issue to avoid delaying completion of this OKR (once work begins).

Status

This was postponed since 2023-Q4 OKR due to higher priority projects like Crescendo migrations and Atree releases.

We can save time and reduce risks by starting this work immediately after Atree Refactoring (while details are fresh):

• Refactor and reduce technical debt #464

[fxamacker]

This was postponed so I can work on atree v0.8.0 (e.g. atree inlining & deduplication, etc.) and also help at onflow/cadence and onflow/flow-go repos with integrations, migrations, tooling, and testing related to atree, etc.

The speed optimizations in this issue would not require spork or migration.

[fxamacker]

This is blocked by

• Refactor and reduce technical debt #464

cc: @j1010001

[fxamacker]

I was asked not to begin work on this several weeks ago because Atree is "efficient enough" (based on profiling report).