Review libflow's interface

[AltraMayor]

Since spinning off f3write in 2020, libflow has become a key aspect of how F3 works and is currently employed in f3write, f3read, f3brew, and f3probe. This pull request streamlines the interface of libflow by

1. Unifying size units with blocks. Each file system and block device has a block unit. Adopting the block unit of the underlying storage minimizes size conversions between units: bytes, sectors, and blocks; and
2. Operating with block order instead of block size. A block size is a power of 2, so operating with its exponent (i.e., $\log_2(\text{block size})$) simplifies arithmetic by replacing multiplications, divisions, and modulos with bit operations: left shift (<<), right shift (>>), and AND (&).

In addition, this interface review brings the following minor improvements:

1. Improved type safety by converting non-negative quantities (e.g., block counts, orders, and sizes) from signed to unsigned integers.
2. Renamed fields and functions to clarify that they operate on blocks rather than bytes (e.g., total_size → total_blocks, step → step_blocks, and get_rem_chunk_size() → get_rem_chunk_blocks()).
3. Dropped outdated func_flush_chunk parameter from init_flow() and the redundant total_processed field from struct flow.
4. Improved accuracy of the reported initial speed in start_measurement().
5. Made dbuf_get_buf() alignment-aware, so several callers dropped their ad hoc solutions for memory alignment.