Validate malformed JNI input in CefRequest/CefPostDataElement (#19, #20, partial #21) - #43

Summary

• CefPostDataElement.setToBytes() with a negative size crashes (signed/unsigned integer bug) #19: N_SetToBytes passed a signed jint size straight into CefPostDataElement::SetToBytes()'s unsigned size_t parameter -- a negative size implicitly became a huge unsigned value, causing a silent buffer over-read/segfault with no diagnostic output at all. Reject size < 0 or size > the actual array length before calling into CEF.
• CefRequest.setHeaderByName() with an empty name crashes the Debug/coverage build (CEF's own CHECK) #20: N_SetHeaderByName already guarded a null Java header name, but a non-null empty string still reached CEF's own CHECK(!name.empty()) (aborts Debug/coverage builds; silently permitted in Release). Reject an empty name too.
• CefRequest.setURL("") and reading back a malformed URL crash the Debug/coverage build (CEF's own CHECK/NOTREACHED) #21 (partial): N_SetURL had the same gap -- guarded null, not empty. Reject an empty URL too. The second, distinct crash CefRequest.setURL("") and reading back a malformed URL crash the Debug/coverage build (CEF's own CHECK/NOTREACHED) #21 describes (NOTREACHED reading back a URL that was setURL()'d with a malformed-but-non-empty string) is not addressed here and remains open.

Ported from coverage/phase1-value-objects-phase2-handlers's 0af4a4e. master had no MalformedInputEdgeCaseTest, so this adds a fresh minimal file with just the three regression tests this fix covers.

Test plan

• Native (ninja jcef) and Java (tools/compile.sh linux64) both build clean.
• Local full-suite verification skipped this round -- the local dev VM is dealing with an unrelated, pre-existing GH Release build: SIGSEGV in libc.so.6 during JVM shutdown after all tests pass #10 shutdown crash that's currently reproducing very reliably on any browser-touching test; relying on this repo's CI (test job) to verify instead.

🤖 Generated with Claude Code