Fix off-by-one in HuffCDIC record iteration for AZW3/MOBI

[Imaclean74]

Fixes #3

Summary

huff_record_count in the MOBI header counts ALL records in the HUFF/CDIC block (1 HUFF + N CDICs). The import loop skips the HUFF record by starting at huff_record_index + 1, but iterates huff_record_count times instead of huff_record_count - 1. This reads one record past the CDIC data, which is typically an image (JPEG) or font record. Since these don't start with CDIC, load_cdic() returns InvalidData("Invalid CDIC header").

Every AZW3/MOBI file using Huffman compression (compression type 0x4448 / 'DH') is affected — all such files fail with Invalid CDIC header on export.

Fix

for i in 0..self.mobi.huff_record_count.saturating_sub(1) in both azw3.rs and mobi.rs.

Concrete example

An AZW3 with huff_record_count=3 and huff_record_index=100:

Record 100: HUFF record (Huffman tables)
Record 101: CDIC record 1
Record 102: CDIC record 2
Record 103: JPEG image  ← bug reads this as CDIC, gets "Invalid CDIC header"

The loop starts at 101 (skipping the HUFF), so it should iterate 3 - 1 = 2 times (records 101–102), not 3 times (101–103).

Verification

Confirmed by four independent implementations of the same format — all subtract 1 from the count:

Implementation	How it handles the count
calibre huffcdic.py	huffs[0] = HUFF, huffs[1:] = CDICs
KindleUnpack mobi_header.py	range(1, huffnum) — loops 1 to count-1
libmobi compression.c	huff_rec_count - 1 CDICs explicitly
SumatraPDF MobiDoc.cpp	cdicsCount = huffmanRecCount - 1

Test plan

• Unit tests added in huffcdic.rs covering valid HUFF/CDIC loading and rejection of non-CDIC data (the exact failure mode)
• Full boko test suite passes (527 tests, 0 failures)
• Tested with a real HuffCDIC-compressed AZW3 (huff_record_count=3, 1 HUFF + 2 CDICs) — opens and exports successfully with fix, fails with Invalid CDIC header without

A stripped AZW3 fixture (47KB, no copyrighted content) is available for manual reproduction — attached in the comment below.

[Imaclean74]

Here's a stripped AZW3 fixture (47KB) for manual reproduction. It was derived from a commercial AZW3 with all text content, images, and metadata removed — only the PDB/MOBI headers, HUFF/CDIC records, and the poison JPEG record (trimmed to 16 bytes) are preserved.

Without the fix, Book::open() succeeds but export() fails when the CDIC loading loop reads the JPEG record at huff_record_index + huff_record_count:

InvalidData("Invalid CDIC header")

[Imaclean74]

huffcdic_stripped.azw3.gz