Skip and log malformed rows during COPY

[ray6080]

Sometimes it's not easy to have a fully cleaned dataset, which could be annoying when it comes to data loading, as users need to repeatedly try through COPY and modify their dataset as we throw errors.
To provide a more smooth COPY experience, it would be good to add an option to COPY statement which allows skipping malformed rows either silently or through logging.
This should cover copying of both node and rel tables.

Notes:

• This issue Rel Table Loading Improvement #1792 is related.
• This feature proposal to Postgres can be a reference, though this proposal never gets implemented inside Postgres.

[ray6080]

The "show warnings" feature in MySQL can be a reference. https://dev.mysql.com/doc/refman/8.0/en/show-warnings.html

[semihsalihoglu-uw]

We can have an IGNORE_ERRORS=true parameter to COPY FROM and LOAD FROM statements.

[ray6080]

To get this feature fully done, we need to support several things:

• Allow ignoring errors inside the CSV reader, which supports LOAD FROM with option IGNORE_ERRORS=true.
• Allow ignoring errors inside casting functions. This needs more work to push casting inside readers.
• Allow ignoring errors like duplicated primary key, missing primary keys from node tables, etc., inside the batch insert pipelines (NodeBatchInsert, IndexLookup).
• Report skipped rows at the end of execution in the query result, including file name, row idx and content of the errored row. (Need a separate fTable inside ResultCollector to collect warning information).

[royi-luo]

#4067 (comment)

• A TODO: we should add a local cache for warnings that we periodically flush to the shared warning cache. This will reduce the amount of lock contention, improving performance.

[royi-luo]

To get this feature fully done, we need to support several things:

* [x]   Allow ignoring errors inside the CSV reader, which supports `LOAD FROM` with option `IGNORE_ERRORS=true`.

* [ ]  Allow ignoring errors inside casting functions. This needs more work to push casting inside readers.

* [ ]  Allow ignoring errors like duplicated primary key, missing primary keys from node tables, etc., inside the batch insert pipelines (NodeBatchInsert, IndexLookup).

* [x]  Report skipped rows at the end of execution in the query result, including file name, row idx and content of the errored row. (Need a separate fTable inside ResultCollector to collect warning information).

Update: a function call show_warnings() was added that reports all cached warnings

[ray6080]

See left todo in #4579.