Honor allow_extra_data=False for unknown-size frames in the C backend - #376

The C backend ignores allow_extra_data=False when a frame omits its
content size and max_output_size exceeds the decoded size.

For example:

import zstandard as zstd

frame = zstd.ZstdCompressor(write_content_size=False).compress(b"foobar")
zstd.ZstdDecompressor().decompress(
    frame + b"junk",
    max_output_size=64,
    allow_extra_data=False,
)

This returns b"foobar" instead of raising ZstdError for the four
unused bytes. The output-buffer resize branch prevents the subsequent
else if trailing-data check from running.

This change makes that check independent of buffer resizing. The default
behavior and explicit allow_extra_data=True remain unchanged.

Errors raised while evaluating the truth value of allow_extra_data are
also propagated correctly, preserving the original exception instead of
returning a result with an exception set and producing SystemError.

Regression coverage includes unknown-size empty and nonempty frames,
exact and spare output capacity, trailing junk and a second frame, and
decoder reuse after rejection. Additional cases cover exceptions from
__bool__ and __len__, with and without trailing data, and verify that
the original exception object is preserved and the decoder remains usable.

Validation on Windows with CPython 3.12.14:

• Of 24 trailing-data subcases, 10 fail against the unmodified C backend;
  all pass with this change.
• All 24 additional truth-value error subcases fail with SystemError
  before the error-handling fix and pass afterward.
• All 48 subcases pass with both the C and CFFI backends.
• The regular test suites pass for both backends, with their existing
  skips.

Related to #181.