Use buffering (bufio?) for reading image tarball on export

[abitrolly]

In #1274 while reading the image tarball from stdin, it is first saved to file, because export algorithm uses random file access as described here #1274 (comment)

To makes the export more efficient, the algorithm could cache image bits in memory until they are no longer needed. So, for example, if manifest.json is parsed, keep the parsed structure in memory and discard cached bytes that contained it. For well-aligned images it will save both speed and memory. For badly aligned the performance will be the same as with current temp file, because temp file is still written into memory tmpfs on Linux.

bufio can potentially help https://pkg.go.dev/bufio but it looks like a lib for helping with string scanning. Not sure it can handle several GB of memory cache efficiently.

And here is a good article with code that allows to see random tar access in debug mode - https://blog.gopheracademy.com/advent-2017/seekable-http/

[github-actions]

This issue is stale because it has been open for 90 days with no
activity. It will automatically close after 30 more days of
inactivity. Keep fresh with the 'lifecycle/frozen' label.

[abitrolly]

Not sure I can reopen this, but I gave it a try without bufio in #1429.