fix(ittage): bound reset ready wait

[FU-max-boop]

Summary

• bound ITTageWrapper.reset() while it waits for io_s1_ready
• expose a configurable 1024-cycle default budget and reject invalid budgets before clocking the DUT
• preserve the existing successful reset sequence and final 10-cycle settling period
• add focused hardware-free tests for immediate readiness, the last allowed cycle, exact timeout behavior, invalid budgets, and constructor forwarding

Problem

ITTageWrapper.reset() deasserted reset and then waited indefinitely for io_s1_ready. If the generated DUT never became ready because of an RTL, build, or configuration problem, the test process could hang without a useful failure.

Implementation

• Count post-deassertion wait cycles and raise TimeoutError with the elapsed count and last observed ready value when the budget is exhausted.
• Accept readiness on the final allowed cycle.
• Keep reset_async() unchanged.
• Move the generated DUTITTage import behind TYPE_CHECKING, allowing the reset control path to be tested without a generated DUT.

The current generated RTL clears five 128-entry tables in parallel and was observed ready 130 cycles after reset deassertion. The configurable 1024-cycle default provides headroom, but is a defensive operational limit rather than a formal protocol bound.

Validation

python -m pytest -p no:cacheprovider -q \
  ut_frontend/bpu/ittage/test/test_reset_timeout.py

• fresh Python 3.12 run: 6 passed
• previous exact-branch generated-DUT ITTAGE run: 11 passed; ready observed after 130 cycles
• git diff --check: passed

Scope boundary

• Production changes affect only the synchronous Python wrapper reset path.
• No RTL or predictor behavior changes.
• The six new tests use a fake DUT and validate timeout/control-flow boundaries, not RTL behavior.