Replace strcmp chains with hashmap lookups

[jserv]

This commit modernizes lexer token recognition by utilizing existing hashmap infrastructure and arena allocation system for performance and consistency, resulting in O(n) → O(1) average case token lookup.

Summary by Bito

This pull request enhances the lexer token recognition system by replacing multiple strcmp chains with efficient hashmap lookups, improving performance from O(n) to O(1). A cleanup function for the lexer hashmaps has also been introduced to manage memory more effectively, ensuring better consistency and performance.

[jserv]

Due to internal error of parser, I can not write elegant code like below:

// Create preprocessor directive lookup table
typedef struct {
    const char *name;
    token_t token;
} directive_entry_t;

static const directive_entry_t directives[] = {
    {"#include", T_cppd_include}, {"#define", T_cppd_define},
    {"#undef", T_cppd_undef}, {"#error", T_cppd_error},
    {"#if", T_cppd_if}, {"#elif", T_cppd_elif},
    {"#ifdef", T_cppd_ifdef}, {"#ifndef", T_cppd_ifndef},
    {"#else", T_cppd_else}, {"#endif", T_cppd_endif},
    {"#pragma", T_cppd_pragma}, {NULL, T_start}
};

// Use binary search instead of linear strcmp chain
token_t lookup_directive(const char *token) {
    int low = 0, high = 10;  // Number of directives - 1
    while (low <= high) {
        int mid = (low + high) / 2;
        int cmp = strcmp(token, directives[mid].name);
        if (cmp == 0) return directives[mid].token;
        else if (cmp < 0) high = mid - 1;
        else low = mid + 1;
    }
    return T_identifier;
}

Instead, I have to specify each item for hash map.