backend: Support OBI buses without rready (#48)

[DanielKellerM]

Closes #48.

Problem

Some OBI configurations omit the optional rready signal (obi_cfg_t.UseRReady = 0), so the subordinate cannot be back-pressured on the R channel. iDMA's OBI read/write managers drive and rely on rready to throttle responses (idma_obi_read.sv:120, idma_obi_write.sv:168), making them incompatible with such buses — exactly as reported in #48.

Approach

Reuse the upstream, already-verified obi_rready_converter (OBI package) rather than reworking iDMA's datapath:

• New backend parameter ObiUseRReady (default 1'b1), threaded idma_backend → idma_transport_layer → the OBI DB template. Emitted only for variants that use OBI.
• ObiUseRReady = 1 → byte-identical to today (direct connect; existing users and tests unaffected).
• ObiUseRReady = 0 → a generate branch inserts obi_rready_converter on each OBI manager port. It buffers responses in a FALL_THROUGH FIFO (depth NumAxInFlight) and credit-throttles the A channel, so the subordinate is never back-pressured, while iDMA's managers keep using rready internally (left completely untouched).
• The converter's obi_a_chan_t/obi_r_chan_t are derived locally via the SV type() operator, so no new type parameters leak into the backend interface (no break for integrators).

Why the converter and not a credit counter inside the iDMA managers: an in-manager credit that returns on buffer push over-issues (the buffer may still hold undrained data, dropping a later response). The converter buffers the raw response before the byte-masking stage, which is correct regardless of buffer occupancy — and it's already verified upstream.

Verification

• make idma_hw_all regenerates all OBI variants; the ObiUseRReady = 1 branch is the previous instantiation verbatim.
• Elaboration (slang, full design via bender script flist-plus):
  • default path: idma_backend_synth_{rw_obi,r_obi_w_axi,r_axi_w_obi} → 0 errors;
  • ObiUseRReady = 0 path (converter live): 0 errors — obi_rready_converter, credit_counter, and the type()-derived channel types all resolve.
• obi_rready_converter/credit_counter are pulled from the existing obi/common_cells deps — no Bender.yml change.

Remaining before merge (why this is a draft)

• Directed simulation against a no-rready subordinate. The OBI backend TB currently bridges the OBI port to AXI (obi2axi_bridge → axi_sim_mem); the proper test threads ObiUseRReady = 0 to the DUT and drives it from obi_sim_mem configured UseRReady = 0 (upstream-supported), checking data integrity + no dropped responses across the simple/medium/error_simple jobs. The converter itself is upstream-verified, but iDMA should prove the integration in sim.

Part of the iDMA 0.7.0 enhancement set; additive and default-off, so it does not gate the release.