Switch from readr to vroom

[peterdesmet]

Readr uses vroom under the hood, so why switch? It has three direct advantages:

1. support for big integers (use vroom::col_big_integer() for integers to avoid overflow? #212)
2. read multiple files directly (rather than calling read function multiple times)

Longer term (especially with Data Package v2 fieldMatch) I think it will be useful to be closer to the vroom dependency.

Reading

• Use vroom::vroom() instead of readr::read_delim()

• Organize arguments vs vroom() not read_delim()

• Don't call vroom::vroom() for each path, but pass it directly

• Run tests

• Update documentation

• Investigate if this applies to vroom (in table-schema.Rmd):

  It relies on readr::read_delim() instead, which interprets line terminator LF and CRLF automatically and does not support CR (used by Classic Mac OS, final release 2001).

• Investigate if this applies to vroom (in table-schema.Rmd):

  since readr::write_csv() used by write_package() formats those to ISO8601, which is considered the default.

Writing

• Use vroom::vroom_write()
• Investigate if this applies to vroom:

  Output is written as UTF-8

• Run tests
• Update documentation

Encoding

• Can we get rid of readr::guess_encoding()?
• Update in data-resource.Rmd

[peterdesmet]

I started switching to vroom, see this diff.

However:

1. readr has readr::guess_encoding() which we use when attaching a file on disk as a resource in add_resource(). While we can try to replicate guess_encoding() as a helper function (which would require the stringi dependency), readr::read_lines_raw() has no equivalent in vroom and cannot easily be replaced by base::readBin(). The latter doesn't support opening gz/zip or remote files. It's a lot easier thus to stick with readr::guess_encoding().
2. vroom::vroom_write() does not guarantee UTF-8 output, while readr::write_csv() does. That is a feature we want.
3. vroom::vroom() bubbles up parsing issues even when silenced (Warning re: parsing problems cannot be suppressed with suppressWarnings() when using altrep tidyverse/vroom#558)
4. readr::read_csv() does support multiple files, so that is not unique to vroom.
5. While not documented, vroom::col_big_integer() can be used in readr::read_csv(). It relies on big64 which is a dependency of vroom. Adding vroom as a dependency to frictionless doesn't increase its size, because it is already a dependency of readr.

Conclusion:

• Stick with readr
• Read multiple files
• Use vroom for col_big_integer (only)

[peterdesmet]

Read multiple files implemented in #281. col_big_integer has a separate issue: #212.