tlv493d: disable test TestTLV493D_ReadContinous

It fails on CI with:
  panic: i2ctest: expected playback to be empty: I/O count 8; expected 9

github actions: update workflows

Use new version that I used in bootstrap and gohci, which hacks for v3
incompatibility with go modules. Will fix this later.

mcp23xxx: fix invalid value; add link to datasheet

See page 12, Table 3-1, and page 17 table 3-5 (bank select is zero on
reset).

bcm283x: golint fix

No functional change.

Made column differential work for SSD1306 (google#444)

Remove go.mod

TL;DR: A project with:
- a valid semver tag equal or above v3.0.0
- with a go.mod file
- but not using versioned imports
is not supported by Go in modules mode; GO111MODULE=on on previous
version, and anything but GO111MODULE=off on likely 1.13 or newer when
the default is switched.

The only way to keep it working simultaneously on versions below 1.9.7
and above 1.11 in go module mode is to remove the go.mod file.

It is currently expected that v4.0.0 will be released with a go.mod
file, which will break Go toolchains before 1.9.7. As a note, the
minimum tested version is currently 1.7.6.

Even then, there's some fun. This is because of the source files using
'// +build periphextra'. The tag 'periphextra' only has effect when
building executables under cmd/ and experimental/cmd/. When set, these
executables call hostextra.Init() from periph.io/x/extra/hostextra
instead of host.Init() from periph.io/x/periph/host. The reason is that
periph.io/x/extra contains all drivers that have external dependencies,
while periph.io/x/periph itself aims to not have any external
dependencies. Yet, it's useful to be able to use, let's say, a FT232H
with cmd/gpio-read. Hence the optional build tag.

The problem is that then go mod complains, because it sees the
periph.io/x/extra/hostextra import, and wants to add it as a dependency
in go.sum. Hello circular dependency! I'm not yet sure how I'll handle
this beside hacking manually the go.sum file or removing this option at
all.

bcm283x: fix invalid aliasing; causing failure on RPi1 and RPi Zero

The regression was introduced in
3d9b008, so version v3.2.0 up to v3.6.0
are affected.

Add smoketest to catch regressions. Manually tested on a RPi1 and RPiZW.

Fixes google#420

mpu9250: heavy doc cleaning.

No functional change but add a lot of TODOs.

fs/sysfs: MIPS: Don't translate all IOCTLs (google#402)

Some IOCTLs are fixed, e.g. i2c-dev. There, we must not translate the
IOCTLs. Instead, the IOCTLs are generated arch-dependent in the fs
package.

- fs: add IO() IOR() IOW() IORW() functions for the equivalent linux macros.
- videocore: Replace ioctl magic with derived value.
  Beside being clearer, the comment in the code was incorrect. The message
  box magic is in fact 'd' = 100 = 0x64, not 0x100.
- videocore: Use dynamic pointer size for ioctl.

i2c: Adds i2c.Addr type and flag.Value interface. (google#362)

* i2c: Adds i2c.Addr type and flag.Value interface.

Add i2c.Addr
Add i2c.Addr Set()
Add Tests.