Add --infer-true/--infer-false for user-defined boolean strings

[johnkerl]

Addresses #1651.

What this does

Adds opt-in flags letting users specify exactly which field values should be inferred as boolean when read from data files:

$ printf 'id,active,verified\n1,True,yes\n2,False,no\n3,true,maybe\n' |   mlr --infer-true True,yes --infer-false False,no --c2p put '$ta = typeof($active); $tv = typeof($verified)'
id active verified ta     tv
1  True   yes      bool   bool
2  False  no       bool   bool
3  true   maybe    string string

This follows option 1 from the issue discussion (inference-level, not a verb), so it works uniformly across CSV/TSV/DKVP/NIDX/etc. input, and also from .mlrrc (infer-true True etc.) per @Poshi's comment.

Approach

• New --infer-true {a,b,c} / --infer-false {a,b,c} flags in the Miscellaneous flag section, alongside the existing inference flags (-S/-A/-O). Each takes a comma-separated list of values.
• pkg/mlrval/mlrval_infer.go: a package-level map[string]bool (nil unless the flags are used, so the default hot path only pays a nil check) is consulted before scan-based inference in inferNormally and inferWithOctalAsInt. -A composes via inferNormally; -S bypasses all inference including this, regardless of flag order.
• Matching is exact and case-sensitive; unlisted values are unaffected. Only inference from data files is affected — JSON string values stay strings (per the issue: JSON booleans are already well-defined), and DSL string literals are unaffected.
• As with ints (0xff) and floats, the original string representation is retained, so CSV/TSV output round-trips unchanged (yes stays yes). JSON output emits canonical true/false — this required fixing marshalJSONBool to use the underlying boolean value rather than mv.String(), which would otherwise emit invalid JSON like bare yes.

Testing

• New unit test TestInferUserDefinedBooleans in pkg/mlrval/mlrval_infer_test.go (go test ./pkg/... passes).
• New regression cases test/cases/io-infer-flags/user-defined-booleans{,-json} (full regression suite passes).
• make lint clean; docs regenerated (reference-main-data-types, flag list, man pages).

Open design questions

• Flag naming/syntax: --infer-true/--infer-false with comma-separated lists mirrors the issue discussion (--true True,on,yes --false False,off,no). Alternatives: a single --infer-booleans True=true,False=false-style mapping flag, or repeatable flags. Also: commas are the separator, so values containing commas can't be specified.
• Case-sensitivity: @aragilar suggested an optional case-insensitivity flag (e.g. --infer-booleans-fold-case). Deferred for now — exact matching keeps the semantics predictable; case variants can be listed explicitly.
• Special values (the "(and special values)" in the issue title, e.g. mapping strings to inf/NaN): deferred, since opt-in inference of the standard true/infinity literals is covered by [feature request] Opt-in type-inferencers for "true", "infinity", etc in file data #965 and could share this mechanism (a user-defined string-to-Mlrval table) if desired.
• DSL exposure: the discussion also floated using the mapping in the boolean() DSL function; currently boolean() is unchanged and only file-input inference is affected.

🤖 Generated with Claude Code