Fix KeyError when spot token index is missing from tokens list

[RWCS-LTD]

Summary

Info.__init__ raises KeyError (or IndexError on pre-dict-refactor versions) and every client crashes at startup when the HL API briefly returns a universe pair whose token index is not present in spot_meta["tokens"].

Reproducer

Observed in production on 2026-05-29: HL added spot pair @367 referencing token index 479 while the tokens list held 464 entries. Every Info() caller crashed for ~54 minutes until HL repaired the response shape.

Quick repro against live API at the time of the incident:

```python
import json, urllib.request
body = json.dumps({"type": "spotMeta"}).encode()
r = urllib.request.urlopen(urllib.request.Request("https://api.hyperliquid.xyz/info", data=body, headers={"Content-Type": "application/json"}), timeout=10)
spot_meta = json.loads(r.read())
n = len(spot_meta["tokens"])
bad = [(u["name"], u["tokens"]) for u in spot_meta["universe"] if max(u["tokens"]) >= n]
print(f"Pairs with out-of-range tokens: {bad}")

Pairs with out-of-range tokens: [('@367', [479, 0])]

```

Fix

Skip universe pairs whose base or quote is not in token_by_index:

```python
if base not in token_by_index or quote not in token_by_index:
continue
```

This lets Info() initialize for all valid pairs and keeps clients alive through any future universe/tokens drift on the API side. New valid pairs come back into scope automatically as soon as their tokens land in the list.

Test

Added `tests/test_info_spot_tokens.py` with a synthetic `spot_meta` that includes one valid pair and one pair pointing at a non-existent token index. The test asserts `Info()` no longer raises and that the valid pair is still registered.

Verified the test fails on `master` without the patch (`KeyError: 5`) and passes with it.

Test plan

• `pytest tests/test_info_spot_tokens.py` passes with the patch.
• Same test fails on `master` without the patch.
• Hand-tested against the live API after HL repaired the response shape — `Info()` initializes correctly and lists 463 of 464 spot tokens (the malformed pair is silently skipped, as intended, until HL adds the missing token).