chore: remove LevelDecoder

[Weijun-H]

Which issue does this PR close?

Closes #2379

Rationale for this change

What changes are included in this PR?

Are there any user-facing changes?

[zeevm]

@tustvold @viirya

Hi,
In my use of parquet-rs, I have a flow that reads and computes over the raw dictionary IDs in data pages, to get to those I'm using LevelDecoder to read repetition and definition levels, this change broke my code.

I'd like to put LevelDecoder back in WDYT?

[tustvold]

LevelDecoder to read repetition and definition levels

What do you think of either:

• Use RleDecoder's directly
• Maintain LevelDecoder within your own project

I would rather avoid maintaining an experimental API that isn't used by any of the first-class APIs

[zeevm]

I am using RleDecoder of course to decode the RLE into dictionary IDs.

I am using LevelDecoder for 2 things:

1. Get the definition levels (so I know which positions are null or defined)
2. Find the position within the ByteBufferPtr where the RLE encoded values start to feed into the RLE decoder in V1 pages since those pages don't have the size of the definition/repetition levels in the page header

[tustvold]

The page header contains the encoding, either Encoding::RLE or Encoding::BIT_PACKED, based on this you can select either to use RleDecoder or BitReader, which should do what you require. This is why LevelDecoder is no longer used, it is redundant.