[Lore0599]

Add cc_stream_downsizer / cc_stream_upsizer

This PR introduces two generic ready/valid stream width converters, parametrized.

New Cells

• cc_stream_downsizer drains one wide beat into WideWidth/NarrowWidth narrow beats, LSB first. It doesn't register the wide word: as long as inp_ready_o is low, the master must hold it stable anyway (standard handshake rule), so each slice is just indexed straight off inp_data_i.

• cc_stream_upsizer does the reverse. Here each narrow beat is genuinely new data, so the first Ratio-1 slices do need to be captured into a register bank. The last slice is the exception: it's streamed out
  combinationally together with the stored ones; in the same cycle it's accepted, instead of being registered and presented a cycle later. That keeps its acceptance directly tied to oup_ready_i, and means the sequence "accept last slice of one word" / "start accepting the first slice of the next" costs zero extra cycles.

HW cost:

Downsizer is register-free besides the counter; upsizer holds (Ratio-1) * NarrowWidth bits of state. Neither adds combinational depth beyond cc_counter's own.

Testbench

Add two testbenches generated by AI 🤖

[colluca]

Few minor comments left

[phsauter]

Assertions are working and testbenches as well, everything looks good.