feat(competition): king-progression timeline from competition.history

[galuis116]

What does this PR do?

summarize_competition (#397) reports the current composite king + score-to-beat, and standings (#387) ranks miners — but nothing tells the story of the competition: who has held the composite crown, in what order, at what score, and how much each new king had to gain to take it. competition.history records exactly that — pr_eval._update_leaderboard appends one entry per crowning, in order.

How

Add trinity.competition_timeline: build_timeline reads competition.history and reports the reign-by-reign progression —

• each king's composite score, per-benchmark breakdown, and gain over the previous king (the delta that had to clear win_margin; the first crown is measured against the 0.0 seed floor);
• the total gain from the first crown to now, and the single biggest leap.

It's read-only and tolerant: a non-dict competition, non-list history, non-dict entry, or an entry missing miner/score are skipped without crashing. scripts/competition_timeline_report.py renders it (text or --json):

| # | king | composite | gain over prev | pr | when |
| 1 | alice (gen 2) | 0.6000 | +0.6000 | #10 | 2026-07-05T00:00:00Z |
| 2 | bob (gen 4)   | 0.8500 | +0.2500 | #20 | 2026-07-06T00:00:00Z |

- current king: bob (0.8500) after 2 crowning(s)
- total gain (first crown → now): +0.2500

Scope

Pure stdlib — no torch, no network. New module + script + tests; touches no existing file. The committed seed leaderboard.json renders "no one has been crowned yet." 10 offline tests cover the gain-over-previous-king math, the biggest leap, the per-benchmark breakdown, the empty/no-competition cases, skipped malformed entries, and load-from-file.

• ruff + mypy clean; pytest tests/test_competition_timeline.py green (10).

Duplication note

New self-contained module; verified no open PRs exist at push time, and it's distinct from standings (current ranking) and summarize_competition (current king) — this is the historical progression. Fourth piece of the competition-object toolkit (after merged #384/#387/#397).

[galuis116]

CI was red because main has unresolved git merge-conflict markers in src/trinity/analysis/__init__.py (a botched merge → the package is invalid Python, so ruff/import fail on every branch). That's unrelated to this PR's files. I've stacked the identical repair here purely to unblock CI — it's the same fix as the in-flight #402/#405 and drops out on rebase once main is green.