Optimizer: cap chain reads of competitor quotes with chain_read_min_secs (optimizer.json, default 600s)

[anderdc]

Summary

Follow-up to #734. That PR added a chain read of competitor quotes (getProgramAccounts per managed direction, 10 Helius credits each, about 21 credits per read) after every re-seed. Only the reads triggered by an API disagreement were rate-limited.

The risk: a re-seed can happen once per 60 s tick. If Helius drops the optimizer's websocket every minute (a throttled or flaky free key is exactly when that happens), the optimizer would do a chain read every minute: about 30k credits a day, over 1M a month from this read alone. That's the whole free tier, and the base miner's swap feed shares the same key.

The change:

• Every chain read now goes through one limit: at most one per chain_read_min_secs. That covers reads after a re-seed, reads triggered by an API disagreement, and retries.
• A read asked for sooner is marked due and taken by the first 5-minute reconcile once the limit allows it.
• Failed attempts count toward the limit, so a failing read is capped too.
• chain_read_min_secs is a new optimizer.json key, default 600, must be >= 0.
• Worst case for this read is now 6 per hour, about 126 credits an hour, even with a feed that drops constantly.
• Normal behaviour is unchanged: real drops are hours apart, so the read after a re-seed still happens immediately.

Tests

• New: test_a_feed_that_drops_every_minute_reads_the_chain_once_per_limit. Reconnects every tick for 16 minutes read the chain twice (at 0 and 10 minutes); with a configured 300 s limit, they read every 5 minutes.
• Updated: a failed read after a re-seed is retried by a reconcile once the limit has passed.
• The config test covers loading chain_read_min_secs and rejecting a negative value.
• pytest: full suite passes.