CDRIVER-5667 Add missing BSON type checks in special key handlers

[eramongodb]

Resolves CDRIVER-5667. Verified by this patch.

Traced the issue to a code pattern discrepancy in _bson_json_read_map_key when handling special keys in the BSON_JSON_IN_BSON_TYPE read state. The BSON type field is reset to BSON_TYPE_EOD (value: 0) when transitioning from BSON_JSON_IN_START_MAP -> BSON_JSON_IN_BSON_TYPE. This PR identified that when subsequently identifying the special key value on re-entry into this function (such as when parsing the key of a subdocument), only some branches performed a check for a pre-existing non-EOD BSON type. Simplified:

if (bson->read_state == BSON_JSON_IN_BSON_TYPE) {
   // Via HANDLE_OPTION expansion.
   if (len == strlen ("$regex") && strncmp ((const char *) val, ("$regex"), len) == 0) {
      if (bson->bson_type && bson->bson_type != BSON_TYPE_REGEX) { // BSON type check. (!)
         _bson_json_read_set_error (...);
         return;
      }
      bson->bson_type = BSON_TYPE_REGEX;
      bson->bson_state = BSON_JSON_LF_REGEX; // LF: "Looking For"
   }

   // Similar pattern for "$options", "$oid", ..., "$numberDecimal".
   else if ...

   // Pattern break: `strcmp` instead of `strlen` + `strncmp`.
   else if (!strcmp ("$timestamp", (const char *) val)) {
      // Pattern break: no BSON type check. (!!)
      bson->bson_type = BSON_TYPE_TIMESTAMP;
      bson->read_state = BSON_JSON_IN_BSON_TYPE_TIMESTAMP_STARTMAP;
   }

   // Similar pattern for "$regularExpression", "$dbPointer", "$code", and "$scope".
   else if ...

   else {
      _bson_json_bad_key_in_type (reader, val);
   }
}

Adding the missing BSON type checks was sufficient to detect and handle the malformed input described in CDRIVER-5667. These checks prevent the state machine from reaching the point where _bson_json_read_append_code is invoked with a precondition violation (the top of JSON stack frame is not a previously-parsed scope document) by detecting+forbidding the presence of any intermediate and unrelated BSON types that could overwrite the current JSON stack frame with their own state.

Some drive-by improvements include:

• Use of an X macro for special key strings to better identify and maintain the list of special keys (e.g. this PR removes several redundant strlen + memcmp calls for duplicate key checks).
• Refactor of the strlen + memcmp (formerly strncmp) pattern into a dedicated macro for consistent reuse.
• Refactor the BSON type check into a macro for consistent reuse.
• Update JSON error test input values with BSON_STR to reduce need for escape characters.

[kevinAlbs]

LGTM.

I am happy to learn "LF" (presumably) means Looking For.

[vector-of-bool]

Good fix and clean-up! LGTM

[vector-of-bool]

Should there be an else before the cast-to-void?

[eramongodb]

I omitted the else because it didn't appear to be necessary. The if is scoped with {}, so there is (AFAIK) no opportunity for else hijacking or other (plausible) mistakes which may produce unexpected behavior at usage sites. A plain ((void)0) seems sufficient to force a semicolon regardless of the macro definition (if or otherwise). If I am incorrect and it is advisable to keep the else anyways, a followup PR can patch it in.

[vector-of-bool]

Since this is so contained, I don't think there's any risk of anything goofy happening. My only thought was about:

if (x)
    HANDLE_OPTION_TYPE_CHECK(...)
else
    lol();

But that would just be a hard compile-error anyway.