I2C: put inline helpers in IRAM for slave ISRs

[Tech-TX]

heavily modified to insure that the 'inline helpers' are truly inline (they are, verified in the dump)

[Tech-TX]

Did you want me to close the other pull request #6894
as this one corrects the same issue in a different way?

The 'inline' attribute on some of the routines seemed to override the 'ICACHE_RAM_ATTR' attribute, and some of the slave ISR routines were ending up in Flash, or were split; part was in Flash, part in IRAM. This caused exceptions when I2C slave events happened during OTA download.

[devyte]

@Tech-TX yes please, let's keep this one.

[Tech-TX]

Latest commit added devyte's comment about not inlining reply(), removed the inline attribute from 2 public functions in the slave side, and moved twi_scl_valley() out of the middle of the slave code and into the master area, as it's a master-only function.

[Tech-TX]

As only one person has seen this error so far, would you mind waiting until Sunday evening or later to merge it? I'd like to run every test against it that I can think of, since the changes touched nearly every lowest-level function. I've eyeballed all of the changes 3 ways, but I'm the eternal pessimist.