Add read-only db-debug agent skill wrapping scripts/db-debug.sh

[TaprootFreak]

What

Adds a read-only db-debug agent skill that wraps the existing scripts/db-debug.sh CLI, so any
agent (and any teammate) gets the documented query modes and database schema without re-explaining
them each session. The script itself is unchanged and stays usable directly on the shell.

The skill follows the open Agent Skills SKILL.md standard and is
tool-neutral: frontmatter is limited to the portable core (name, description), and the body
uses repository-relative paths only.

It also documents three companion balance-forensics scripts (compare-balance-logs.sh,
inspect-asset-balance.sh, sum-asset-balances.sh) as part of the same skill.

Why fold them in rather than make a second skill: all three use the byte-identical
DEBUG_ADDRESS/DEBUG_SIGNATURE/.env auth and the same /gs/debug endpoint as db-debug.sh,
and operate on the same log / FinancialDataLog balancesTotal snapshot structure that
reference.md already documents. A separate "balance-forensics" skill would split the same
interpretation rules across two places; documenting them as modes of db-debug keeps the knowledge
in one location. This directly serves the recurring total-balance-step investigations.

Why this location

.claude/ and CLAUDE.md are git-ignored in this repo, so shared, version-controlled skills live
in a neutral top-level skills/ directory instead. Each developer points their own agent at it
(per-developer agent config stays local). This is documented in a new Agent skills section in
CONTRIBUTING.md.

Files

• skills/db-debug/SKILL.md — modes, invocation, and the liquidity_balance snapshot query
• skills/db-debug/reference.md — /gs/debug semantics, FinancialDataLog / balancesTotal, and
  the liquidity_balance schema (loaded on demand)
• CONTRIBUTING.md — new Agent skills section

Notes

• Read-only only: the skill documents SELECT-only usage; /gs/debug already enforces this
  server-side through SQL AST validation.
• No secrets committed — DEBUG credentials stay in each developer's local .env.
• An optional pre-execution guard that blocks write statements is left as a possible follow-up
  (defense in depth).

PR completeness

• Migration: N/A (no entity/column changes)
• Environment / Infrastructure: N/A
• Service updates: N/A (no DTO/interface changes)
• Frontend synchronization: N/A (no API contract changes)

Closes #3955. Related: #3956 (db-debug.sh efficiency improvements).