fix: add integer overflow and memory safety guards to C parser layer

[billdenney]

• rc_dup_str (shared.c): make ptrdiff_t→int narrowing explicit; add bounds check so diff<0 or diff>INT_MAX calls Rf_error() instead of silently truncating. Add thread-safety comment documenting that the global parser state is intentionally not mutex-protected (R is single-threaded).

• sbuf.c: add signed-integer overflow guards before each size-arithmetic expression in sAppendN, sAppend, and addLine (both the string-buffer and line-pointer-array growth paths), preventing R_Realloc from receiving a negative or near-zero size.

• parseSyntaxErrors.h (getLine): change col accumulator from int to size_t and add an explicit INT_MAX guard, preventing R_Calloc from receiving a wrapped-negative size when a source line exceeds INT_MAX bytes.

• All 13 trans_*() parser entry-points: check strlen(gBuf)>INT_MAX before casting to int for the dparse() length argument; R strings are capped at INT_MAX-1 bytes by R itself, so this guard protects against direct C-level misuse.

• tests/testthat/test-memory-safety.R: two regression tests (no skip) confirming normal-sized inputs still parse correctly; three skip() tests documenting the >2 GB boundary conditions with strrep()-based allocations (avoiding paste0+rep() intermediate vectors that caused memory exhaustion).

• NEWS.md: add v0.0.7 entry.

[billdenney]

I'm breaking this into smaller parts for easier review. I intend to close this when the separated PRs are generated.