NAIF DAF mutation should be its own struct again

[ChristopherRabotin]

High level description

To mutate a DAF in a previous version (< 0.7?) one had to call to_mutable, which unlocked mutation functions. This would copy the underlying bytes to allow for mutation. This was removed in favor of a single DAF which allowed mutation.

With hindsight, that was a mistake.

The bytes were stored in a Bytes type, which is zero copy from the allocation point (or static &[u8]). Now, everything is stored in a BytesMut, which requires a Copy of the underlying bytes on initialization. A Clone operation is also a deep copy instead of a shallow pointer copy like Bytes.

That's a problem because the header of each file is read in the Almanac::load function to determine the file type, and if it's a DAF kind, the whole byte array is copied again to read the DAF. If the DAF was a Bytes then it would not need to copy the data again: the loading would be zerocopy.

This issue shows up in a flamegraph:

The penalty is only paid once, but it should be paid zero times.

Requirements

• Create a MutDAF struct, which has a freeze function that would return to an immutable DAF (or to_immutable()?)
• The existing DAF functions that use a &mut self should be moved to the MutDAF struct.

Test plans

The existing tests should call my_daf.to_mutable() before trying to modify anything, and then freeze() before loading said file back in the Almanac.

Upload the flamegraph.svg from the built cargo test --release -- flamegraph and flamegraph -- ../target/release/.. -- flamegraph and the flamegraph command to show that the copy no longer happens. This proves the fix is correct.

Design

This design worked well before.