Add weighted fair scheduling for vqueues via scope rules

[PascalThesing]

Problem

With experimental-enable-vqueues, dispatch capacity is proportional to a
service's queued key count: a service with 100K queued keys starves few-keyed
services (measured: ~50 batching keys at 0.05% dispatch share).

Changes

1. Two-level WRR: the eligibility ring and the invoker-concurrency waiter
   list rotate over scheduling groups (invocation scope, else service name);
   weight N = N slots per cycle regardless of key count. Per-queue DRR inside
   a group unchanged; service groups fixed at weight 1.
2. Weights as rules: restate rules set <scope> --weight N
   (UserLimits.scheduling_weight, shown in rules list / sys_rules).
   With experimental-enable-scope-inheritance, ctx.call children inherit
   the caller's scope — one rule governs the whole call chain.

Validation

115 tests (exact weighted ratios, work conservation, wire compatibility);
benchmark: one weight-10 rule → +52% throughput for the flooding service
while the starved batchers flush twice as often.

[github-actions]

All contributors have signed the CLA ✍️ ✅
_{Posted by the CLA Assistant Lite bot.}

[PascalThesing]

I have read the CLA Document and I hereby sign the CLA