Roll out codegen registry to all client classes (#114)

[martinkersner]

Closes #114

Converts remaining client methods off hardcoded /api/v… paths onto self.request_endpoint("<op_id>", **wire_params), single-sourcing path/method/path-query split/defaults from the generated _endpoints.py.

Validation policy (settled)

Each class keeps its own semantic + required checks unchanged (check_required_parameter(s), market ∈ [spot,futures], sort ∈ [asc,desc], currency/type enums, "no data found", DataFrame shaping). request_endpoint only owns path/method/split/defaults + a spec-driven required backstop. No error messages or behavior changed; existing tests pass unmodified.

Converted (fully)

• forex, naver, telegram
• cex_fee, cex_ticker, cex_wallet_status
• cex_announcement
• liquidation (all), open_interest (all)
• funding_rate: history, exchanges, symbols
• cex_symbol: metadata, tags, oi, oi_stats, liquidation
• premium: exchanges

Left on hardcoded paths — registry/openapi.yaml gaps (would break pinned wire contracts)

• funding_rate.latest — registry funding_rate_latest lacks sort+limit (class supports them).
• cex_token.updates — registry cex_token_updates lacks sort (test pins sort=desc on wire).
• cex_symbol.cautions / delistings — registry lacks base (class sends it).
• cex_symbol.volume — registry cex_symbol_volume lacks exchange (test pins exchange on wire).
• premium.__call__ — registry premium uses snake_case wire keys (source_exchange) but class/test pin camelCase (sourceExchange), and registry lacks ~40 min/max/funding filter params the class sends.

These need upstream datamaxi-codegen / openapi.yaml fixes before conversion; filed as follow-ups.

Test plan

• pytest -m "not integration": 113 passed, 11 skipped, 116 deselected.
• black --check + flake8 on datamaxi/ + tests/: clean.
• grep for /api/v in datamaxi/**.py (excluding _endpoints.py + doc-link comments/URLs): only the 5 registry-gap methods above remain.