Add multi-issue and inclusive-range CLI inputs

[riddim-developer-bot]

Summary

Two related additions to the OSS `llm-cost` CLI:

1. Multiple positional issue IDs:
   ```bash
   llm-cost EPAC-1940 EPAC-1921 FAC-67
   ```

2. Inclusive ranges as a single positional:
   ```bash
   llm-cost EPAC-1990-1999 # 10 issues, EPAC-1990..EPAC-1999
   llm-cost EPAC-1990-1999 FAC-60-70 EPAC-1940 # mix ranges + singletons
   ```

The range syntax is unambiguous because Linear issue keys are strictly `-` (no other hyphens are permitted), so the second hyphen can only mean range. Single ranges are capped at 10,000 expanded IDs.

Output

When more than one ID is requested, the output is a per-issue summary table plus a TOTAL row. From real data on this machine:

```
COST ROLLUP — EPAC-1920-1925 (source: usage.jsonl)
6 issues requested, 4 had data

Issue Sessions Turns Tokens API cost
─────────────────────────────────────────────────────────
EPAC-1920 1 57 9.5M $6.79
EPAC-1921 1 3,048 476.0M $292.37
EPAC-1922 1 159 23.9M $19.20
EPAC-1923 1 1,718 266.7M $162.14
─────────────────────────────────────────────────────────
TOTAL 4 4,982 776.1M $480.51
(skipped: EPAC-1924, EPAC-1925 — no sessions)
```

Single-issue behavior (the full per-bucket / quota / pricing block) is unchanged.

New library API

```js
import {
computeMultiIssueRollup,
expandIssueArg,
expandAllIssueArgs,
} from 'llm-cost-attribution';

expandIssueArg('EPAC-1990-1999')
// → ['EPAC-1990', 'EPAC-1991', ..., 'EPAC-1999']

const { ids, requestedCount } = expandAllIssueArgs(['EPAC-1990-1999', 'EPAC-1995']);
// ids.length === 10, requestedCount === 11 (range had 10 + 1 singleton, dedup)

const multi = await computeMultiIssueRollup(['EPAC-1990-1999'], loadOne);
// → { label, positionals, requestedIds, requestedCount, issues, missing, totals }
```

The loader is pluggable so the same function works against raw transcripts OR a usage.jsonl source — the CLI passes whichever loader matches `--from-usage`.

Drive-by fix

`attachPricingToRollup` previously took `totals.models[0]` (alphabetical sort) for the rate lookup, which is wrong when Claude includes `` markers (session-restart placeholders that sort before real model names). Now iterates the model list until it finds a name the rate table knows. FAC-67 now correctly prices at $9.15 instead of showing `—`.

Test plan

• 66 of 66 tests pass via `node --test` (was 51; +15 new multi-issue tests covering expansion, dedup, range bounds, mixed inputs, missing-issue separation, label generation, and aggregate pricing)
• `node --check` clean on every .mjs
• Range expansion verified against real data: `llm-cost EPAC-1920-1925 --from-usage ` → 4 of 6 have data, totals $480.51
• Mixed positionals verified: `llm-cost EPAC-1940 FAC-67 WEB-175 --from-usage ` → totals $100.11
• Single-issue invocation produces byte-identical output to the previous version

What's NOT in this PR

Per the scoping conversation, this PR does only multi-issue + range. Not added:

• `--team` flag (sugar for "all EPAC-* issues")
• `--repo` flag (filter by workspacePath)
• Linear project-level rollup (would need a tracker integration; out of scope for the OSS package)