Cleanup keyutils types and api.rs

[josephlr]

This patchset makes many changes in preparation for fixing issue #24 (to no longer need libkeyutils). Doing everything in one PR would be very large, so this PR is just all of the internal cleanup that doesn't change the external API of keyutils.

Specifically:

• libkeyutils-sys is moved to keyutils-raw (and version is updated to match keyutils)
• Adds KEY_TYPE_KEYRING and KEY_TYPE_BIG_KEY (as they are in the kernel documentation).
• As the key/keyring ID can never be zero, the type of a key id is changed from i32 to NonZeroI32.
  • Thus, some parameters have to be Option<NonZeroI32>, but this is FFI safe per Allow null-pointer-optimized enums in FFI if their underlying representation is FFI safe rust-lang/rust#60300
• This makes Rust 1.34 keyutils's minimal version.
• In keyutils-raw: change key_serial_t to just be KeyringSerial, which is then just reexported by keyutils.
• In keyutils-raw: change key_perm_t to just be KeyPermissions, which is then just reexported by keyutils.
• Remove KEY_REQKEY_DEFL_* constants. Instead, DefaultKeyring in keyutils-raw just has enum values with these constants (it is also just reexported to keyutils).
• Cleanup internal helper methods in api.rs to use the new types. I also simplified the internal *_impl and check_ methods. This is the largest part of the PR, but is mostly just a find/replace.

Again, note that we continue linking libkeyutils, and this only ends up with a very tiny change to the external API:

• DefaultKeyring:
  • no longer has a serial() method
  • Implements TryFrom<libc::c_long> instead of From<i32>

[mathstuf]

Thanks for all the work this past week! Once this is rebased, I'll take a closer look at it now that the other PRs have been merged.

[josephlr]

Thanks for all the work this past week! Once this is rebased, I'll take a closer look at it now that the other PRs have been merged.

Done, and let me know if you want to split this up. I've also revised the PR's description to be more accurate.

[mathstuf]

Just worried about kernel upgrades sliding in panics without changing the code when a new default keyring value is added. Looks great otherwise.

[mathstuf]

Thanks!