Allow enabling serde/std without also requiring serde_cbor/std to be enabled

[kylefleming]

Currently serde_cbor's std feature must be enabled if serde/std is also enabled, even if serde/std is enabled separately by another crate somewhere else in the dependency tree.

However, this doesn't necessarily need to be the case. serde provides a StdError type in order to get around this (see serde-rs/serde#1620). This is a type provided by serde as a re-export of std::error::Error when serde/std is enabled and as an export of a custom serde::std_error::Error type with the same interface as std::error::Error when serde/std is not enabled. By conforming to serde::ser::StdError instead of std::error::Error, serde_cbor can rely on the std/no_std state of serde rather than requiring that both libraries' std feature be either both enabled or both disabled.

This allows serde_cbor to be in no_std even if serde/std is enabled. While this might not make a ton of sense at first glance, this change would lead to a much more robust dependency mechanism when serde and serde_cbor are involved, by allowing a crate to depend on serde_cbor with default features disabled and not have to worry if another crate (maybe even higher up in the dependency graph) enables serde/std completely independently. Ideally, the ultimate goal is that features are additive and not dependent on the state of other features laterally in the dependency graph.

This PR does 2 things:

• Changes the type of the std error that serde_cbor::Error conforms to from std::error::Error to serde::ser::StdError.
• Adds a line to .travis.yml to test building with serde/std enabled and serde_cbor's default features disabled in order to test that this configuration is buildable.

[stevenroose]

This LGTM!

[stevenroose]

I'm not sure who's in charge atm for merging in this repo. But I think this is a good change.

[rjwalters]

it seems that only @pyfisch has admin rights

[kylefleming]

This was auto-closed completely unintentionally.