Bug fixes, features and some public API changes

[citreae535]

Bug fixes and refactors

1. Breaking: change Service.delete to borrow rather than consume self

The underlying win32 API call merely marks the service for deletion. It can be called even if the service is not stopped and will not invalidate the caller's open handle to it. Making the function not consuming self allows the caller to decide when to close the handle. Docs and examples are updated to reflect this change.

2. Refactor double_nul_terminated implementation

• Rename from_vec to from_slice since it takes a slice, not a Vec.
• Allocate WideString in from_slice with pre-calculated capacity.
• Add unit tests for from_slice.

  3. Bump windows-sys to 0.45

  An additional feature Win32_Storage_FileSystem is pulled in just for the flag FileSystem::DELETE. This does not feel right, but hard-coding it ourselves feels worse.

New Features

4. Add public method Service.set_delayed_auto_start

This function configures if an auto-start service should be delayed.

5. Add public method Error.is_os_error

This function checks if an error from win32 API call matches a specific OS error code.

Public API changes

These three are more opinionated public API changes based on my own needs, If these do not feel right for you, I'll revert them.

6. Breaking: remove ServiceDependency

• It does not seem to do anything useful. The '+' prefix of a group dependency has to be passed to the win32 API. An OsString will suffice here since it is up to users to distinguish between service and group dependencies.
• If users want to check if a dependency is a group, they can simply check if the OsString starts with '+'.

  7. Breaking: change ServiceInfo to a borrowed type

  This change allows the caller to retain individual ownership of the fields of ServiceInfo without having to clone them.

  8. Possibly breaking: replace err-derive with thiserror

  • Both crates derive std::error::Error for an enum with slightly different syntax. For this crate it is a drop-in replacement and unlikely to break anything.
  • thiserror has some extra features and are better maintained.
  • Because thiserror is vastly more popular (76,032,688 total downloads and 8400 dependent crates compared to err-derive's 2,175,409 and 71), it is very likely that many users of this crate already depend on thiserror and will have less dependencies if this switch is made. (In fact, crates.io shows that this crate is the most downloaded dependent of err-derive)

---

This change is 

[pronebird]

Hi,

Thanks for your efforts. However, your PR introduces a lot of changes, some of them are good, others are breaking or subject to discussion. It would be much simpler to review your changes if they were somehow isolated one from the other via separate PRs.

[citreae535]

Close in favor of separate PRs