It cannot select the compression tool while reading from stdin

[Vincenzo Romano]

#System:
ArchLinux (up-to-date) on x86_64
tar: GNU tar 1.35
gzip 1.13

# Steps to reproduce:
1. Get/create a compressed archive (e.g. "tar cfz test.tar.gz /etc/xdg")
2. Check it can be read and handled by tar (e.g. "tar tvf test.tar.gz")
3. Try to handle it from stdin (e.g. "tar tvf - < test.tar.gz")

# Actual result:
tar cannot select the proper compression tool/library and thus cannot
be used in a shell pipeline.
$ tar tvf - < test.tar.gz
tar: Archive is compressed. Use -z option
tar: Error is not recoverable: exiting now

# Expected result:
Same as step no.2

# Possible Rational
TAR normally reads from the archive file a few blocks to understand
whether any compression tool/library is needed. It likely "rewinds"
the read pointer to the beginning and enables the needed tool/library.
TAR doesn't follow the same path when reading from STDIN. I can
imagine preview bytes are not pushed back in the buffers for later
processing. But this can be done in a number of different ways like
moving the preview bytes in the buffer to be used by the handling
functions.
The statements in the documentation chapter 8.1.1 (Creating and
Reading Compressed Archives) can be considered
incorrect when not false: you never read directly from any device or
pipeline as you likely place those bytes in some memory buffer for later
handling. Smart management of those buffer can lead to the solution of
the problem.

# Disclaimer
I could not yet find enough time to analyze current source code.