Inline ESP::getCycleCount() to make it safe to call from ISRs

[dok-net]

As the subject line says, it has been found by users (of EspSoftwareSerial), that the function is not ISR-safe right now. Inlining should efficiently fix this, IMHO.

[me-no-dev]

to make it safe, mark it as IRAM ;) now it is not safe, because it has to be loaded from flash. ESP32 != ESP8266

[dok-net]

Is that an observation you can explain based on generated code? What part of inline, __attribute__((always_inline)), and inline assembly still allows for flash access here, when invoked from a code that resided in IRAM? I don't get it - what's the difference to ESP8266 then?

[dok-net]

@me-no-dev I've looked at the output from objdump, seems like the code in question is always inlined. Whenever it's not (by changing the attributes, removing inline, moving the function into Esp.cpp, whatever), I get errors from the compiler from calling getCycleCount() in ISRs. To make a long story short, this PR expresses what the compiler optimizations affect right now anyway, and that's a good thing for the human reader, isn't it?

[me-no-dev]

And how about you mark it as IRAM and make it easily understandable that it can be run in ISR?

[dok-net]

Done.