Refactor BMP row decoding to work on byte slices

[telecos]

Refactors BMP row decoding functions to separate I/O operations from pixel processing logic using pre-allocated buffers.

Changes

Uncompressed and bitfield formats (read_full_byte_pixel_data, read_16_bit_pixel_data, read_32_bit_pixel_data):

• Allocate a single row buffer outside the with_rows closure
• Read entire rows with one read_exact() call per row
• Process pixels from the buffer without additional I/O

RLE-compressed formats (read_rle_data):

• Introduced RleReader struct with 8KB internal buffer
• Reads compressed data in chunks to minimize system calls
• Provides read_byte() and read_bytes() methods for RLE instruction parsing

read_palettized_pixel_data already uses this pattern (row buffer allocated outside closure).

This changeset is going towards #2696