Write file-level statistics when writing ORC files with zero rows

[vuule]

Description

Fixes #14675
Write file-level statistics even when stripe-level statistics don't exist (no stripes).
Written statistics are in line with Pandas - zero sum, no min/max.

Checklist

• I am familiar with the Contributing Guidelines.
• New or existing tests cover these changes.
• The documentation is up to date with these changes.

[mythrocks]

I can't think of a better name for this function. I fear the name implies that it's the close() that's skipped.
skip_footer_write_on_close() is too long.

If we can't think of a succinct name, maybe we should keep skip_close().

[vuule]

omg, my updated code for this function got lost between commits.

I think skip_close is a pretty accurate name, given that we raise the _closed flag in skip_close and close() returns immediately if that flag is raised.

I'll update the docs above, sorry for the confusion.

[mythrocks]

LGTM!

[ttnghia]

Thanks for working on this.
Here is the latest output, which seems to be correct now:

File Version: 0.12 with FUTURE by Unknown(5) 
Rows: 0
Compression: SNAPPY
Compression size: 262144
Calendar: Julian/Gregorian
Type: struct<id:bigint,p:bigint>

Stripe Statistics:

File Statistics:
  Column 0: count: 0 hasNull: false
  Column 1: count: 0 hasNull: false sum: 0
  Column 2: count: 0 hasNull: false sum: 0

Stripes:

File length: 94 bytes
Padding length: 0 bytes
Padding ratio: 0%

[ttnghia]

Wow mutable lambda is great.

[vuule]

FWIW, I was told it's a code smell (but this is test code)
I think in theory mutable lambda means we should use an STL algorithm instead

[vuule]

/merge