feat: skip range-checking PUSH operations in KERNEL mode

[Nashtare]

Following internal discussion, as part of our assumptions are the KERNEL validity, as source of truth, we do not need to range-check the hardcoded PUSH operations happening in context 0. This involves a new CpuGeneralColumnsView to hold a temporary variable to account for the increased degree when filtering the associated CTL.

This allows to massively reduce both BytePackingStark and MemoryStark trace lengths, as PUSH operations normally require a range-check on all bytes of the pushed value.

Example on an ERC20 transfer:

• BP: 17023 -> 537
• Memory: 363111 -> 232806

This PR also cleans-up the way BytePackingOps are appended, by skipping empty sequences which would be filtered out after anyway.

Idea initially suggested by @4l0n50.

[Nashtare]

I'm going to update the specs too, it's missing a mention of the new view