Use an improved semantic validation layered on top of the existing
resolver (around ~40 LoC) instead of its own dedicated logic
(~240 LoC).  Not only the improved type-checker is leaner and faster,
it's also more accurate in some corner cases.
The type-checking process is also deferred to the compilation stage,
instead of earlier in the parsing stage.

* Enable Postgres driver to use secrets for configuring SSL parameters

Adding secret related properties for SSL options to Postgres driver

Move `conn-props->secret-props-by-name` to secret.clj since it's needed directly there, too

Add `us-east-2-bundle.pem` to `test-resources/certificates` for testing Postgres with SSL connectivity to our RDS instance (and a README.md explaining how it differs from the existing `ssl` directory)

Updating `value->file!` to have better logic for building the error message when the existing file is not found (for better UX from the Database admin page)

Updating frontend to support the `visible-if` value being an array, in which case any value may match the current details value, in order for that field to be visible

Adding secret related properties for SSL options to Postgres driver

Updating CircleCI config.yml to refer to the RDS instance when running Postgres SSL test

Implement server side expansion of transitive visible-if dependency chain

Update shouldShowEngineProvidedField on client to consider multiple key/value pairs in the visible-if map, and only show the field if ALL are true

Adding new test to confirm transitive visible-if resolution, and cycle detection

Add Cypress test for SSL field visibility

* restyle the dataset education modal

* restructure the modal code and styling

* add t tags to strings

* move Popover to subfolder

* Add TippyPopover component

* Add popover theme styling

* Add/use EventSandbox component

* make TippyPopover lazy

* improve demo page

* add unit tests for TippyPopover

* let most defaulted props be overridable

* rename cypress helper fn

* clean up .info example

* Add tests for buildCollectionTree

* Rename test objects for better readability

* Allow filtering collection tree items by models

* Hide collections without datasets in dataset picker

* Keep parent nodes if they don't have target models

* Pass real dataset instance to QB components

* Add basic support for "dataset" QB mode

* Add simple DatasetEditor page

* Fix dataset editor turns into ad-hoc mode on any change

* Navigate back from the editor

* Use `getQuestion` selector for datasets too

* Allow removing "Visualize" button from Notebook

* Allow hiding native query editor's parameters list

* Add `isInitiallyOpen` prop to native query editor

* Add "Edit query" button to question sidebar

* Add ResizableNotebook component

* Add query editor to dataset editor

* Fix native dataset updates change QB mode to "view"

* Hide NavBar on dataset editor page

* Add ability to save updated dataset query

* Extend Cypress helper to create native datasets

* Add E2E tests for dataset query editor

* Use ActionButton for "Save changes" button

* Reset query changes on "Cancel" button click

* Move DatasetEditor to QB components

* Fix QB selectors unit tests

* Test `getIsResultsDirty` tests for datasets

* Test `getQuestion` QB selector

* Lock dataset display to table in editor

* Allow to disable visualization click handlers

* Disable table click actions for dataset editor

* Fix can't open native query sidebars

* Remove not used component

* Fix scroll effect when resizing notebook

* Fix "back" button behavior

* Fix missing translation

* Fix template tags assignment in dataset QB mode

* add funnel static viz

* prettier

* support funnel formatting options, change measureText signature

* fix tests

* fix types

* unhardcode legend text style

* decrease static funnel step font size, fix warnings

* update specs

Field references in custom expression must be correctly resolved as
dimensions, segments, or metrics. For instance, `[Rating] > 4 AND
[Expensive]` means that [Rating] is a dimension and [Expensive] is a
segment.

Before this change, this process was carried out inside the parser. This
led to various corner cases which were not handled well, e.g.
`COALESCE(CASE([X], [Y]))` apparently resolved both [X] and [Y] as
dimensions. This was incorrect, as [X] has to be a segment, since it's
part of the conditional clause in CASE().

With this change, all field references are resolved as an additional
pass. Thus, the resolver has the full visibility of the operand types
for both unary and binary operators, as well as argument types of all
functions, including variadic ones such as CONCAT, COALESCE, and CASE.