v0.3.0

v0.3.0 — Karnaugh-map layout (logic_kmap)

logic_minimize tells you the answer; logic_kmap now shows you the map it came from. The new tool
lays the function out on Gray-coded axes (1–4 variables, cells 1 / x / 0) and maps every term
of the minimal cover back to the cells it circles — so a hand grouping can be reproduced and checked
instead of trusted:

logic_kmap { expr: "a & !b | a & c" }
K-map of a & !b | a & c over 3 variable(s) (rows a | cols b, c)
    00 01 11 10
0 |  0  0  0  0
1 |  1  1  1  0
minimal cover: a & !b | a & c
  a & !b: 2 cell(s) — r1c0(m4) r1c1(m5) [essential, contiguous]
  a & c: 2 cell(s) — r1c1(m5) r1c2(m7) [essential, contiguous]

• Groups are the cover, not the catalogue: each group is one term of the returned minimal cover,
  with row / col / minterm for every cell, an essential flag, and a wraps flag for groups that
  leave one edge and re-enter on the opposite one (that is why !c is a single 4-cell group).
• Don't-cares are drawn, not hidden: dontCare cells appear as x and groups may circle them
  (a & !b & c | a & !b & !c with dontCare [6, 7] collapses to a over 4 cells). A dontCare entry
  that is already true is drawn as 1 and the result says so.
• Honest limits: 5+ variables are refused by name (a map grid stops being readable) with a pointer to
  logic_minimize; a contradiction is an all-zero map with no groups; a tautology is one all-free group.
• logic_minimize also reports coverTerms — which prime implicants actually form the answer, in answer
  order — so the two tools can be cross-checked term by term.

Tests 144 → 168, coverage 96.32% → 96.95% (baseline ratcheted up). The in-repo oracle
(test/oracle/anchors.py) gained a K-map section that lays the map out independently in Python plus
--check invariants: every minterm appears exactly once, groups only circle cells inside on ∪ dc,
the groups cover the on-set, a wrapping group's cell count matches its cube, and Gray order changes
exactly one bit per step including the wrap.

---

v0.3.0 — 卡诺图布局（logic_kmap）

logic_minimize 给答案，logic_kmap 现在把答案画回图上：新工具把函数铺到格雷码坐标（1–4 变量，
格子里是 1 / x / 0），并把最小覆盖的每一项反查回它圈到的格子——手画的圈可以被复现和核对，
而不是只能凭信任。

• 圈的是覆盖，不是目录：每组对应返回覆盖里的一项，逐格给出 row/col/minterm，附 essential
  标记；跨边界折返的组标 wraps（!c 在三变量下就是一个 4 格组）。
• 无关项画出来、不藏起来：dontCare 画成 x，圈组可以把它们圈进来；本来就已为真的无关项画成 1
  并在结果里说明。
• 限制如实报：5 变量以上直接点名拒绝（网格不再可读）并指向 logic_minimize；矛盾式是全 0 图、
  无圈组；重言式是一个覆盖全图的组。
• logic_minimize 同时给出 coverTerms（到底用了哪几项素蕴含项，按答案顺序），两个工具可逐项对账。

测试 144 → 168，覆盖率 96.32% → 96.95%（基线已上抬）。随仓 oracle 增加卡诺图区（Python 独立布局）
与 --check 不变量：每个最小项恰好出现一次、圈组只圈 on ∪ dc 内的格子、圈组覆盖全部最小项、
折返组的格数等于其立方体大小、格雷码每一步（含折返）只变一位。