feat(sv56): variable bit-width support (8-32 bit) for speech voltmeter

[jr2804]

Summary

Extends the P.56 speech voltmeter (sv56) module to support variable bit-width input files from 8 to 32 bits per sample, up from the previous 16-bit limitation.

Changes

Core algorithm (sv-p56.c, sv-p56.h)

• Added int bitno parameter to init_speech_voltmeter() controlling threshold count and peak detectors
• New constants: SVP56_MAX_NO_BITS (32) and SVP56_MAX_THRESHOLDS (31) for array sizing
• State struct gains thres_no and bitno fields; arrays expanded accordingly

Variable-width I/O (sv56-util.h — new)

• sv56_bytes_per_sample(bitno) — returns 1/2/3/4 bytes based on bit depth
• sv56_raw2fl() — converts raw bytes to normalized float (-1..+1) in native byte order
• sv56_fl2raw() — converts normalized float back to raw bytes with truncation and clipping
• Supports 8-bit (1 B), 16-bit (2 B), 24-bit (3 B), 32-bit (4 B) storage; legacy 12-15 bit masking preserved

CLI tools (sv56demo.c, actlevel.c)

• Replaced short buffer[] with unsigned char raw_buf[] for variable-width I/O
• Replaced sh2fl()/fl2sh() with sv56_raw2fl()/sv56_fl2raw()
• Fixed out-of-bounds mask[] array access for bitno > 16
• bitno parameter now controls both P.56 algorithm AND file format

New utilities

• sv56test.c — minimal non-interactive test program for automated verification
• sv56-convert.c — utility to convert between bit-width formats (used for test data generation)

Tests

• 6 new CTest cases: run + verify for 16-bit, 24-bit, and 32-bit inputs
• All 11 sv56 tests pass (5 original + 6 new)
• Full test suite: 1173/1182 pass (9 pre-existing failures in unrelated modules)

Documentation (doc/manual/sv56.tex)

• Updated init_speech_voltmeter signature and parameter docs
• Documented new constants, struct fields, and variable-width I/O
• Updated example code

Build system

• CMake minimum version updated to 3.10
• Added sv56test and sv56-convert build targets

Backward compatibility

• bitno=16: Identical behavior to previous version (2 bytes/sample, divide by 32768, native byte order)
• bitno=12-15: Same 2-byte file format, correct normalization and LSB masking
• bitno=8, 24, 32: New functionality
• All existing test reference files produce identical results

[ludomal]

@jr2804 - Thanks for the great contribution. Would you mind committing the test reference files?

[jr2804]

Thanks for the review! All feedback addressed in the latest two commits:

• .gitignore: Removed the blanket src/**/test_data/ rule. Added a local src/sv56/.gitignore instead to scope ignore rules to the sv56 module only.
• Test reference files: Committed in 353558a — includes voice_24.src, voice_32.src, and .ref files for 16/24/32-bit verification.
• CMakeLists.txt: Noted — happy to align with whatever cmake version normalization you settle on for the new release.