perf: replace O(n²) concat with push in metaculus fetchMarkets

[realfishsam]

Fixes #551

[realfishsam]

PR Review: PASS (NOT VERIFIED)

What This Does

Replaces all = all.concat(results) with all.push(...results) in the metaculus fetchMarkets.ts pagination loop, avoiding O(n^2) array re-allocation on each page.

Blast Radius

• Core: core/src/exchanges/metaculus/fetchMarkets.ts only
• No OpenAPI, SDK, or type changes

Findings

1. The all variable is declared with let (line 45) but is no longer reassigned after this change. It could be tightened to const for clarity, though this is non-blocking.
2. With push(...results) and BATCH_SIZE = 100, each spread is well under the V8 call-stack limit (~65k args). No risk of "Maximum call stack size exceeded" at this batch size.
3. Correct behavior preserved: all.length continues to reflect cumulative results for the early-exit check on line 64.

PMXT Pipeline Check

• Field propagation: N/A
• OpenAPI sync: N/A
• Type safety: OK

Semver Impact

patch -- internal perf improvement, no API or behavioral change

Risk

Not verified against live Metaculus data. The change is mechanically safe but untested in this review.