Release 4.8.0: expression DSL v2, scoped contexts, docs overhaul

[iskandr]

Summary

Rolls up all changes since 4.7.0 into a single release.

Expression DSL v2 (PR #103)

• Scoped contexts: wt.affinity.score, shuffled.affinity.score, self.affinity.score — reserved keywords replace the old WT() wrapper
• len: peptide length from precomputed column, works with scopes (wt.len)
• count('C'): dynamic amino acid counting from peptide string, works with scopes (wt.count('KR'))
• __repr__ round-trips: all expression objects serialize to parseable strings
• Scope class: wt, shuffled, self_scope module-level instances for Python API
• Parser: wt, shuffled, self are reserved context keywords
• Removed: WT class (clean break, no backward compat)

CLI expression parity (PR #101, included in #103)

• parse_expr() — full recursive-descent parser for --rank-by expressions
• All transforms, aggregations, arithmetic, method qualification work from CLI strings
• Fixed _build_ranking_strategy routing for dotted expressions
• Switched to predictors_from_args (mhctools >= 3.4.0)

Docs overhaul

• README rewritten for cohesion: removed siloed sections, integrated examples showing DSL features
• Removed all "Python-only" claims — everything works via CLI strings
• Removed fictional advanced example (referenced non-existent predict_column)
• Consolidated predictor tables, updated for mhctools 3.4.0 models
• Updated deploy.sh and RELEASING.md

Design docs

• docs/expression-semantics.md — full specification of scope/prefix model, grammar, internal representation
• Expression data design: RNA expression, variant evidence (isovar), single-cell data as first-class DSL fields
• Isovar integration path: peptide source + evidence annotations

Tests

• 540 tests (was 397), all passing on Python 3.10/3.11/3.12

Test plan

• 540 tests pass locally
• CI passes

[coveralls]

coverage: 88.779%. remained the same — release-4.8.0 into master