Merge release to master

* needs triage label to simplify triage

* cause obviously this is how multiple labels should be specified

* Full support for all dimension types in expression editor (field-id, field-literal, fk->, joined-field, expression)

* Fix various bugs in metabase-lib

* Limit 'card-has-ambiguous-columns?' to native queries + use real DB id in composeThisQuery

* Fix and add to tests

* Swap Question constructor metadata/card args + misc other cleanup

* Refactor and fix drill-throughs to support explicit joins etc

* Misc FE metadata code cleanup (#11044)

* Fix joined dimension options, e.x. custom field from a saved query

* Cleanup schema_metadata and related method names

* Use shorter aggregate/filter/breakout/sort/join methods everywhere

* Add UnderlyingRecordsDrill tests

Merge master to release 0.33.x

add an endpoint so that users who don't own a pulse can unsubscribe themselves from the pulse

* Validate i18n at compile time; fix busted format strings 

* Lint fixes 

* Delegate getList to getObject

* Use getMetadata for metadata entities getObject selectors

* Per object caching in getObject using re-reselect

* clean up and reorder install steps [ci skip]

* correct the upload-your-code instructions [ci skip]

* Correct suggested value for "availability"

* Fix bullets and quoting

* Add periods at end of sentences

* Fix grammar and capitalization

* Specify config blocks that options are from

* Align emphasis of "NOTE" lines

* Fix more bullets

* Fix one more NOTE

* some small tweaks and screenshot updates [ci skip]

The current text implies that a password must be _exactly_
`complexity.total` length, whereas that is actually the minimum length.

* Don't fail entire query if we don't know how to convert result to literal

* Code cleanup 

* Full support for all dimension types in expression editor (field-id, field-literal, fk->, joined-field, expression)

* Fix various bugs in metabase-lib

* Limit 'card-has-ambiguous-columns?' to native queries + use real DB id in composeThisQuery

* replace metadata.database[...] etc with metadata.database(...)

* Fix and add to tests

* Revert "replace metadata.database[...] etc with metadata.database(...)"

This reverts commit 0f5aa69be184a1a767675e253eace47a419d970a.

* Fix joined dimension options, e.x. custom field from a saved query

Fixes #10523.

* remove Google Fonts usage

* local Lato in resources/frontend_client/app/fonts

* create assets/fonts aliases, update publicPath, use absolute paths in css

* add svg to extensions

* update flow config to account for frontend_client assets

* add explicit .svg extension to no_results usage

* remove google fonts references from CSP headers

Merge `release-0.33.x` into `master`

* remove extra padding on confirm modal

* add modal examples to help keep an eye on things

c3p0 has an option to test connections when fetched from the connection pool intended to fix that exact problem, so this was a simple matter of enabling it. It is not enabled by default because it adds overhead to connection checkout, especially for JDBC drivers that don't support the full JDBC 4 API, which includes a Connection.isValid() method. However, all of our JDBC drivers are JDBC 4 compliant, meaning c3p0 can check connection validity fairly efficiently.

With the data warehouse on the same machine as the Metabase server, the added overhead was in the order of ~100µs, or an order of magnitude or two below the threshold of human perception. With the data warehouse in AWS us-east-1 and the Metabase server in San Francisco, the overhead was around ~70ms, which is basically typical network latency for such a request.

Based on those results, it appears that the additional cost of this test is roughly equal to the network latency of the request. IRL the Metabase server and data warehouse are likely to be located in closer geographical proximity to one another than my trans-contintental tests, if not in the same AWS region (or equivalent). I would expect the added overhead for most query executions for most setups to be in the 1-10ms range and the p99 to be under 100ms extra latency.

Accepting that overhead seems like a no-brainer when considering that an entire class of bugs will be fixed by enabling the option.