* don't ignore-optional on yarn install for the ui dep ci job

* update circle yaml

* ci: re-enable browserstack tests ...

- Also sets "strict mode" bash with -euo pipefail at executor level.

* ci: Makefile: more compatible SHELL var

* ci: remove some trailing whitespace

* ci: remove redundant set -eux -o pipefail

- In some cases, just replace with -x where debug output might be useful.
- In most cases, don't bother with -x, it is a secret leak risk, and
  circleci prints the command body before executing it anyway.

* try specifying yarn version

* set yarn version

* Revert "try specifying yarn version"

This reverts commit 6b4e57a304d999d72d4b07468a6b4a30880f1f4c.

* ci: break config into separate files

* Untangle githooks

* githooks: fix whitespace

* .hooks/pre-commit: add ui -> lint-staged check

- We no longer require dependency on husky with this change.

* ui: remove husky dependency and config

- The previous commit obviates the need for it.
- We will now have to manage these hooks by hand, but this removes
  the conflict between husky-installed hooks and those in the .hooks dir.

* ui: update yarn.lock with husky removed

* .hooks/pre-commit: always use subshell + docs

- Always use subshell means we consistently exit from the
  same place which feels less complex.
- Docs are necessary for horrible bash like this I think...

* Makefile: remove old husky githooks

- Husky has installed a handler for every single git hook.
- This causes warnings on every git operation.
- Eventually we can remove this, but better not to confuse
  people with these messages for now.

* ci: fix go build tags

* Makefile: improve compatibility of rm call

- Looks like the xargs in Travis does something different to the one
  on my mac, this more verbose call should be safe everywhere.

* ci: fix make target names

* ci: fix test-ui invocation

* Makefile: simplify husky hook cleanup

* ci: more focussed readme