Minimize header use, move Ticker function definitions into cpp file

[dok-net]

This is a corresponding port of ESP32 review results.

[devyte]

In the current code, all methods that take a "seconds" arg are routed through a single _attach_s() which is where the float multiplication is done. This is to have the multiplication in a single place, because on the ESP float arithmetic is done in sw.
In other words, the current code pays the price of an additional forwarder method to reduce the duplication of the generated code for the float multiplication.
On the other hand, you moved the methods out of the header.
Please check size of the bins for current code vs. this PR for a case that uses many of the methods.

[dok-net]

@devyte Are you certain that float multiplication is even inlined?
I think the changes I've now made reduce ROM usage across the board - one ticker, two tickers, same mode, two different modes (attach_ms_scheduled, once).

[devyte]

@dok-net I see you reverted the move from h to cpp. Did you compare the bin size lf your current proposal vs. unifying the float mult in a single attach method as is currently the case?

[dok-net]

@devyte My comment in the ESP32 PR was more informative: better than 204 bytes saved by inlining in header

As to your precise question, I don't see the issue you're hinting at, as far as I can tell from objdump either, the multiplication is not unrolled/inlined, therefore it's rather insignificant in how many places a 1000UL * x or 1000000UL * x expression exists. This is from extensive trying out all sorts of combinations of inline vs. cpp impl, nested functions vs. copy&paste. The version I have in this PR is the only one that turned out well.

That said, I don't expect any single sketch to use all modes of the Ticker at once, YMMV if you do. But for using two Ticker instances, one with attach_ms, the other with once_scheduled, this is what I base my findings on and that's what this PR is quite good at.

[devyte]

@dok-net the relevant code is in effect only when using attach() with a float arg.

[dok-net]

@devyte Naturally, I've tested all sort of combinations, also on ESP32, where attach_ms() scales to µs (1000ULL *) :-)

[dok-net]

@devyte CI is broken - not by me (?)

[earlephilhower]

The pleasures of DevOps and needing sources from 15 different projects. The Python version in the Win VM was changed, and installed into a different dir than before. Fixed now, hopefully, permanently.

[devyte]

Ticker.cpp was moved, so the diff is visible.
ticker.h was not moved, it was (D)eleted in one place and (A)dded in the new one. Therefore, the diff can't be seen here in the PR, and instead the file is fully red in the origin and fully green in the destination.
Please fix that, or redo the PR, so that I can review.

[dok-net]

@devyte It's not on purpose, git doesn't recognize it as a move, the internal difference probably is to large. Could you do a local diff by hand, please?