narrow is_stdin() to documented fileno() failure modes

[HrachShah]

The bare except Exception in is_stdin() (httpie/uploads.py) was too broad: it would swallow every error that fileno() could raise, including ones that signal real bugs in the function itself (NameError, MemoryError, etc.) and ones that no caller expects is_stdin() to handle. KeyboardInterrupt and SystemExit both inherit from BaseException so they still propagate, but the policy in this codebase (and the policy followed in several other recent PRs) is to narrow broad handlers to the documented failure modes of the underlying call.

The actual documented failure modes for file.fileno() on a file-like object are:

• AttributeError — object has no fileno attribute at all (plain BytesIO, custom streams, anything that doesn't expose a real fd).
• io.UnsupportedOperation — stream types like StringIO that explicitly disallow fileno(). This is a subclass of OSError, so catching OSError transitively covers it.
• ValueError — "I/O operation on closed file" for closed file objects (a closed tempfile, a closed BytesIO, etc.).
• OSError — low-level errors on a corrupted/closed fd (EBADF, EINVAL, etc.).

Narrowed the handler to (OSError, ValueError, AttributeError) and kept the same return False semantics for the not-stdin case. Added a short comment listing the four documented failure modes.

Added a TestIsStdin class to tests/test_uploads.py with five unit tests: no fileno attribute, closed file, StringIO, BytesIO, and a real os.pipe() fd that is not stdin. The last test patches sys.stdin via monkeypatch with a surrogate so the comparison side is deterministic; pytest's own capture machinery replaces sys.stdin with a DontReadFromInput whose fileno() raises io.UnsupportedOperation, which would otherwise make the test depend on the runner.