Experimental: Strongly typed types of time

[lemunozm]

Description

The idea of this PR is to have Millis and Seconds as different types for the compiler. This helps to never mix by error different duration units and improve the readability of some parts of the codebase.

To achieve this, we need:

• A wrapper type over integers that allows us different integer-like types to tell the compiler not to mix types (check num_wrapper.rs file).
• Some explicit conversions from seconds to millis that will change the value accordingly
• Implement everything from BaseArithmetic to be able to use this types as normal u64 (i.e. give to pallet_timestamp's Moment the new Millis type.

How times should be used now in the code base?

• If the pallet just carries a time unit, without knowing what unit is, use a generic Moment (probably most of the times a <<T as Config>::Time as Time>::Moment), as usual. You can use the trait TimeUnit to get the time as the unit you want.
• If the pallet uses a moment type that specifically represents millis, seconds or days, you should use cfg_primitives::{Millis, Seconds, Days, ..}. There is no need for an associated type in the pallet.

Blockers

It seems like parity_scale_codec doesn't allow to implement HasCompact trait required by BasicArithmetic:

• Custom types with HasCompact paritytech/parity-scale-codec#613

  • unlock this work if they allow us to implement HasCompact

• Remove HasCompact from BaseArithmetic paritytech/polkadot-sdk#5004

  • unlock this work if they do not force us to use HasCompact

• Found a solution: last message of Custom types with HasCompact paritytech/parity-scale-codec#613

Remainders

• Requirement::DelayTime uses a u32 to hold seconds when it should be a u64 wrapped into Seconds type. Nevertheless, this requires a migration in an storage of pool-system.

[lemunozm]

FooU64 and BarU64 are different types for the compiler, and you can implement different traits to them. In memory and for serde and codec there are just plain u64