dtls: redefine `dtls_write` without `static inline`

[JKRhb]

While https://github.com/eclipse/tinydtls/pull/159 made a very useful addition by introducing dtls_writev, it made dtls_write unusable in shared libraries by turning it into a static inline function. This PR proposes reintroducing dtls_write as a "regular" function when not building for an RTOS (Contiki, RIOT, or Zephyr), making it available in shared libraries again.

Using a doxygen member group, it should be possible to reuse the documentation for both function variants, reducing redundancy.

[boaks]

FMPOV, if a static inline function is not working, I would just replace it by a non-static variant without #ifdef .

(My C summit was 30 years ago, so I don't know, if "static inline" works for shared libraries as well and just requires some compiler switches.)

[mrdeep1]

I think this is more to do with backwards binary API comparability. You have a compiled application using a .so shared library. Library is updated, but application is not re-complied. Application fails because dtls_write() is no longer there.

[boaks]

My 2 cents.

About this function:

• if we go back, then I would prefer without ifdef.

About backwards binary API compatibility in general:

• this is 0.8.6 ... not 1.5.0
• in my experience, the very most are using it for embedded as static link library
• "backwards binary API compatibility" requires in my opinion also stable data structs in the API. Especially introducing new functionality e.g. selectable cipher suites (CCM/CCM8) in my opinion naturally goes into dtls_handler_t, and will then require to keep the old API as it is and additional add new ones. Not that this is not impossible, but FMPOV, I prefer to spend my time in other topics, which have a better pay off.

[boaks]

LGTM