fix(pdf): keep the minus sign on losses in standard-font PDFs (#1982) (#1987)

Intl sv-SE formats negatives with U+2212, which the bundled react-pdf Helvetica/Courier fonts cannot render, so a loss printed as a profit. lib/pdf/number-text.ts pdfNumberText maps U+2212 to an ASCII hyphen and prints negative zero unsigned; routed through financial-statement, kassaflodesanalys, reskontra, momsdeklaration, payslip (fmt, the literal Preliminar skatt sign and the calculation formula text) and operational-report templates, with content-stream regression tests. The K2/K3 arsredovisning templates keep main's formatPdfKronor from #2013.

Closes #1982

tests/pdf-text.ts

@@ -0,0 +1,47 @@
+import { inflateSync } from 'node:zlib'
+
+/**
+ * Text a rendered @react-pdf/renderer document actually carries, as one
+ * string per page content stream (test helper).
+ *
+ * Standard-font text is written as WinAnsi bytes inside `[...] TJ` / `(...) Tj`
+ * operators. Inflating the streams and decoding those operands byte-for-byte
+ * shows what a viewer will draw, which is the only way to prove a glyph made
+ * it into the file: a character with no glyph in the font (U+2212 in
+ * Helvetica, issue #1982) survives as an unmapped control byte and is drawn
+ * as nothing. Thousands separators come back as U+00A0 and are normalised to
+ * a plain space so assertions can use "4 684".
+ */
+export function pdfTextStrings(pdf: Buffer): string[] {
+  const raw = pdf.toString('latin1')
+  const out: string[] = []
+  const streamRe = /stream\r?\n/g
+  let match: RegExpExecArray | null
+  while ((match = streamRe.exec(raw)) !== null) {
+    const start = match.index + match[0].length
+    const end = raw.indexOf('endstream', start)
+    if (end === -1) continue
+    const bytes = Buffer.from(raw.slice(start, end), 'latin1')
+    let content: string
+    try {
+      content = inflateSync(bytes).toString('latin1')
+    } catch {
+      content = bytes.toString('latin1')
+    }
+    // Only page content streams carry text objects; embedded font programs
+    // and images are binary and would produce false hex/paren matches.
+    if (!/\bBT\b[\s\S]*\bET\b/.test(content)) continue
+    const parts: string[] = []
+    const opRe = /<([0-9a-fA-F]+)>|\(((?:\\.|[^\\)])*)\)/g
+    let op: RegExpExecArray | null
+    while ((op = opRe.exec(content)) !== null) {
+      if (op[1] !== undefined) {
+        parts.push(Buffer.from(op[1], 'hex').toString('latin1'))
+      } else {
+        parts.push(op[2].replace(/\\([()\\])/g, '$1'))
+      }
+    }
+    if (parts.length > 0) out.push(parts.join('').replaceAll(' ', ' '))
+  }
+  return out
+}