refactor(protocol): rename _wait_for_block to _wait_live

[QuiteYellow]

@mbillow proposed this rename in #54 and offered to drop it if I preferred the old name. I would rather have it, so it lands here on its own.

Nothing in the helper is block-specific: it waits on an Event in slices and polls _check_live() between them, so a dead reader surfaces as SessionClosedError within one poll interval. That is true on main today, and #54 adds post() as a second caller, at which point _wait_for_block reads as the wrong name.

Why separately

#36 rewrote this function in the same lines #54 renames, and the merge has a trap. #54's side of the conflict ends on:

        deadline = time.time() + per_wait

while #36 moved the function to time.monotonic(). The loop body below merges cleanly and still reads deadline - time.monotonic(), so the mismatch never appears inside the conflict markers. Resolving it the obvious way mixes epoch seconds with uptime seconds, slice_s pins to the poll interval forever, and the per-attempt timeout stops existing: the wait then ends only when the response arrives or the reader dies, so a silent device hangs the call and the retransmission never fires. Every test stubs this function out, so nothing catches it.

Landing the rename as a rebase onto #36 keeps time.monotonic() by construction, and #54 rebases onto a tree that already has the new name.

Contents

The function, its one call site in _exchange_block, and the six stub lines across the three test files. The docstring and the _BLOCK_LIVENESS_POLL_S comment are @mbillow's wording from #54, taken verbatim so those hunks disappear from their diff on rebase.

403 pass. Ruff reports the same count on the touched files before and after.